release: v0.3.5

v0.3.0-alpha.0

.cargo/config.toml

@@ -0,0 +1,5 @@
+[build]
+rustflags = ["-C", "target-cpu=native"]
+
+[alias]
+dev = "run -p brk_cli --features brk_server/bindgen"

.claude/prompts/changelog.md

@@ -1,137 +0,0 @@
-# Changelog Generation Prompt
-
-Update docs/CHANGELOG.md for ALL latest releases not present in the file. Use ONLY git commands - no other sources.
-
-## MANDATORY PROCESS - FOLLOW EXACTLY - NO EXCEPTIONS:
-1. Run `git tag --list --sort=version:refname` to get releases in order
-2. Process EXACTLY ONE release at a time
-3. For EACH SINGLE release: run `git diff [previous-tag]..[current-tag]`
-4. **MANDATORY ANALYSIS STEP**: Before writing ANY changelog entry, you MUST:
-   - Analyze each file change and explain what the code is doing
-   - Identify the purpose and impact of each modification
-   - Group related changes together logically
-   - State clearly what functionality is being added, removed, or modified
-   - If you cannot understand what a change does from the diff, explicitly say so
-5. Only AFTER completing the analysis, write the detailed changelog entry
-6. Update the CHANGELOG.md file with that ONE entry
-7. STOP. Ask me if you should continue to the next release.
-
-## CRITICAL CONSTRAINTS:
-- **NEVER EVER** process multiple releases in one go, even if there are many
-- **NEVER** say "let me continue more efficiently by processing multiple releases"
-- **NEVER** batch releases together for any reason
-- If you feel tempted to process multiple releases, **STOP** and process only one
-- Context window concerns do **NOT** justify batching - process one release only
-
-## ABSOLUTE REQUIREMENTS:
-- **NEVER** read commit messages, PR descriptions, existing changelog, or any text documentation
-- Use **ONLY** the actual code changes shown in git diff output
-- Process releases **ONE BY ONE** - I don't care if there are 100 releases
-- **MANDATORY**: Before writing changelog entries, demonstrate understanding by analyzing what each code change accomplishes
-- Be **HIGHLY DESCRIPTIVE** about what each code change does and why it matters
-- Don't be conservative - write detailed explanations of the impact and purpose of changes
-- **If you don't understand a change from the code diff alone, DO NOT GUESS - say so explicitly**
-
-## SOURCE OF TRUTH:
-- `git diff` output is the **ONLY** source of truth
-- If you can't determine what a change does from the code diff alone, say so explicitly
-- Ignore **ALL** text/documentation - focus purely on code additions, deletions, and modifications
-
-## CHANGELOG FILE REQUIREMENTS:
-- Add a header at the top of the CHANGELOG.md file: `<!-- This changelog was generated by Claude Code -->`
-- Ensure this header appears before any changelog entries
-
-## CHANGELOG WRITING RULES:
-
-### RELEASE TITLE FORMAT:
-**MUST** use this exact format: `## [vX.Y.Z](https://github.com/bitcoinresearchkit/brk/releases/tag/vX.Y.Z) - YYYY-MM-DD`
-Use the actual release date from git tag information
-
-### ABSOLUTELY FORBIDDEN PATTERNS:
-- **NEVER** mention line counts (e.g., "with 138 lines", "1,290 lines removed")
-- **NEVER** use vague action words: "Enhanced", "Improved", "Updated", "Expanded", "Restructured", "Refactored", "Modified", "Adjusted"
-- **NEVER** write sections about Cargo.lock or dependency updates unless they represent major functional changes
-- **NEVER** use the format "Action: File with vague description"
-- **NEVER** mention version bumps of local crates (e.g., "Updated all crate versions from 0.0.61 to 0.0.62") - this is implied by the release version
-- **NEVER** mention dependency version changes in external crates unless they enable new functionality visible in the code
-- **NEVER** write entries like "Updated dependencies" or "Cargo.lock maintenance"
-
-### REQUIRED WRITING STYLE:
-- Write what the code **actually DOES**, not that it was "enhanced" or "improved"
-- Be **specific about functionality**: "Added transaction validation logic", "Implemented caching for API responses"
-- Focus on **business/functional impact**: "Enables users to...", "Fixes issue where...", "Adds support for..."
-- **Mandatory structure**: Group by: Breaking Changes, New Features, Bug Fixes, Internal Changes
-- Include GitHub file links for major changes (max 5 per entry)
-- **Skip entirely**: minor dependency bumps, Cargo.lock changes, and local crate version bumps
-
-### MANDATORY ANALYSIS WORKFLOW:
-**BEFORE writing any changelog entry, you MUST:**
-
-1. **Code Comprehension Check**: Go through each modified file and explain:
-   - What specific functionality is being added/removed/changed
-   - What the new/modified functions/structs/methods do
-   - How the changes affect the overall system behavior
-
-2. **Impact Assessment**: For each change, determine:
-   - Is this a new feature, bug fix, breaking change, or internal improvement?
-   - What user-facing or system behavior changes result from this code?
-   - What problem does this change solve?
-
-3. **Logical Grouping**: Organize related changes together:
-   - Group files that work together to implement a single feature
-   - Separate breaking changes from additions
-   - Distinguish user-facing changes from internal refactoring
-
-4. **Understanding Verification**: Before writing changelog text, state:
-   - "I understand this change does X because the code shows Y"
-   - If unclear: "I cannot determine the purpose of this change from the diff alone"
-
-**ONLY AFTER completing this analysis should you write the changelog entry.**
-
-### WHAT TO FOCUS ON (IN ORDER OF PRIORITY):
-1. **New functionality** - What can users now do that they couldn't before?
-2. **Breaking changes** - What existing functionality changed or was removed?
-3. **Bug fixes** - What specific problems were resolved?
-4. **Internal changes** - New modules, significant refactoring, architecture changes
-5. **Skip completely** - Dependency updates, version bumps, Cargo.lock changes
-
-### VERBOSITY REQUIREMENTS:
-- **Minimum 3-4 bullet points per section** when changes exist
-- **Each bullet point should be 1-2 sentences** explaining both what changed and why it matters
-- **For new features**: Explain what the feature does and what problem it solves
-- **For bug fixes**: Describe the problem that was fixed (inferred from the code changes)
-- **For internal changes**: Explain the architectural or structural improvement
-
-### EXAMPLES OF GOOD vs BAD:
-
-#### ❌ BAD EXAMPLES:
-- "Enhanced: Chain analysis with sophisticated blockchain processing capabilities"
-- "Updated: brk_rolldown from 0.0.1 to 0.1.0 with comprehensive bundling improvements"
-- "Version Bump: Updated all crate versions from 0.0.61 to 0.0.62"
-- "Improved error handling"
-- "Refactored codebase"
-- "Updated dependencies"
-
-#### ✅ GOOD EXAMPLES WITH ANALYSIS:
-
-**Analysis**: "Looking at the diff, I see a new `TransactionAnalyzer` struct was added with methods `calculate_fee()` and `is_coinbase()`. The struct takes transaction data and provides analysis methods. This enables users to programmatically analyze transaction properties."
-
-**Changelog**: "Added new `TransactionAnalyzer` struct that provides methods for computing transaction fees and detecting coinbase transactions"
-
-**Analysis**: "The diff shows error handling was added around block parsing where previously there was an unwrap(). Now it returns a Result and handles the empty block case explicitly. This prevents panics when processing malformed blocks."
-
-**Changelog**: "Fixed panic when processing blocks with zero transactions by adding explicit empty block handling and proper error propagation"
-
-**Analysis**: "I see a new caching layer was implemented with a HashMap storing block hashes as keys and block data as values. The API endpoints now check this cache before making network requests. This should improve performance for repeated queries."
-
-**Changelog**: "Implemented new caching layer for blockchain queries, reducing API response time by storing frequently accessed block data in memory"
-
-#### ❌ BAD EXAMPLES (NO UNDERSTANDING):
-- "Enhanced error handling" (What specific errors? How were they enhanced?)
-- "Improved performance" (What was improved? How?)
-- "Updated transaction logic" (What specific logic? What changed?)
-
-## FINAL REMINDER:
-**PROCESS ONLY ONE RELEASE. THEN STOP AND WAIT FOR MY CONFIRMATION.**
-
-You must be thorough and verbose - if there are code changes, there should be substantial changelog content explaining what those changes accomplish.

.dockerignore

@@ -0,0 +1,3 @@
+.git
+target
+docker

.github/workflows/release.yml

@@ -1,296 +0,0 @@
-# This file was autogenerated by dist: https://axodotdev.github.io/cargo-dist
-#
-# Copyright 2022-2024, axodotdev
-# SPDX-License-Identifier: MIT or Apache-2.0
-#
-# CI that:
-#
-# * checks for a Git Tag that looks like a release
-# * builds artifacts with dist (archives, installers, hashes)
-# * uploads those artifacts to temporary workflow zip
-# * on success, uploads the artifacts to a GitHub Release
-#
-# Note that the GitHub Release will be created with a generated
-# title/body based on your changelogs.
-
-name: Release
-permissions:
-  "contents": "write"
-
-# This task will run whenever you push a git tag that looks like a version
-# like "1.0.0", "v0.1.0-prerelease.1", "my-app/0.1.0", "releases/v1.0.0", etc.
-# Various formats will be parsed into a VERSION and an optional PACKAGE_NAME, where
-# PACKAGE_NAME must be the name of a Cargo package in your workspace, and VERSION
-# must be a Cargo-style SemVer Version (must have at least major.minor.patch).
-#
-# If PACKAGE_NAME is specified, then the announcement will be for that
-# package (erroring out if it doesn't have the given version or isn't dist-able).
-#
-# If PACKAGE_NAME isn't specified, then the announcement will be for all
-# (dist-able) packages in the workspace with that version (this mode is
-# intended for workspaces with only one dist-able package, or with all dist-able
-# packages versioned/released in lockstep).
-#
-# If you push multiple tags at once, separate instances of this workflow will
-# spin up, creating an independent announcement for each one. However, GitHub
-# will hard limit this to 3 tags per commit, as it will assume more tags is a
-# mistake.
-#
-# If there's a prerelease-style suffix to the version, then the release(s)
-# will be marked as a prerelease.
-on:
-  pull_request:
-  push:
-    tags:
-      - '**[0-9]+.[0-9]+.[0-9]+*'
-
-jobs:
-  # Run 'dist plan' (or host) to determine what tasks we need to do
-  plan:
-    runs-on: "ubuntu-22.04"
-    outputs:
-      val: ${{ steps.plan.outputs.manifest }}
-      tag: ${{ !github.event.pull_request && github.ref_name || '' }}
-      tag-flag: ${{ !github.event.pull_request && format('--tag={0}', github.ref_name) || '' }}
-      publishing: ${{ !github.event.pull_request }}
-    env:
-      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          persist-credentials: false
-          submodules: recursive
-      - name: Install dist
-        # we specify bash to get pipefail; it guards against the `curl` command
-        # failing. otherwise `sh` won't catch that `curl` returned non-0
-        shell: bash
-        run: "curl --proto '=https' --tlsv1.2 -LsSf https://github.com/axodotdev/cargo-dist/releases/download/v0.30.0/cargo-dist-installer.sh | sh"
-      - name: Cache dist
-        uses: actions/upload-artifact@v4
-        with:
-          name: cargo-dist-cache
-          path: ~/.cargo/bin/dist
-      # sure would be cool if github gave us proper conditionals...
-      # so here's a doubly-nested ternary-via-truthiness to try to provide the best possible
-      # functionality based on whether this is a pull_request, and whether it's from a fork.
-      # (PRs run on the *source* but secrets are usually on the *target* -- that's *good*
-      # but also really annoying to build CI around when it needs secrets to work right.)
-      - id: plan
-        run: |
-          dist ${{ (!github.event.pull_request && format('host --steps=create --tag={0}', github.ref_name)) || 'plan' }} --output-format=json > plan-dist-manifest.json
-          echo "dist ran successfully"
-          cat plan-dist-manifest.json
-          echo "manifest=$(jq -c "." plan-dist-manifest.json)" >> "$GITHUB_OUTPUT"
-      - name: "Upload dist-manifest.json"
-        uses: actions/upload-artifact@v4
-        with:
-          name: artifacts-plan-dist-manifest
-          path: plan-dist-manifest.json
-
-  # Build and packages all the platform-specific things
-  build-local-artifacts:
-    name: build-local-artifacts (${{ join(matrix.targets, ', ') }})
-    # Let the initial task tell us to not run (currently very blunt)
-    needs:
-      - plan
-    if: ${{ fromJson(needs.plan.outputs.val).ci.github.artifacts_matrix.include != null && (needs.plan.outputs.publishing == 'true' || fromJson(needs.plan.outputs.val).ci.github.pr_run_mode == 'upload') }}
-    strategy:
-      fail-fast: false
-      # Target platforms/runners are computed by dist in create-release.
-      # Each member of the matrix has the following arguments:
-      #
-      # - runner: the github runner
-      # - dist-args: cli flags to pass to dist
-      # - install-dist: expression to run to install dist on the runner
-      #
-      # Typically there will be:
-      # - 1 "global" task that builds universal installers
-      # - N "local" tasks that build each platform's binaries and platform-specific installers
-      matrix: ${{ fromJson(needs.plan.outputs.val).ci.github.artifacts_matrix }}
-    runs-on: ${{ matrix.runner }}
-    container: ${{ matrix.container && matrix.container.image || null }}
-    env:
-      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-      BUILD_MANIFEST_NAME: target/distrib/${{ join(matrix.targets, '-') }}-dist-manifest.json
-    steps:
-      - name: enable windows longpaths
-        run: |
-          git config --global core.longpaths true
-      - uses: actions/checkout@v4
-        with:
-          persist-credentials: false
-          submodules: recursive
-      - name: Install Rust non-interactively if not already installed
-        if: ${{ matrix.container }}
-        run: |
-          if ! command -v cargo > /dev/null 2>&1; then
-            curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
-            echo "$HOME/.cargo/bin" >> $GITHUB_PATH
-          fi
-      - name: Install dist
-        run: ${{ matrix.install_dist.run }}
-      # Get the dist-manifest
-      - name: Fetch local artifacts
-        uses: actions/download-artifact@v4
-        with:
-          pattern: artifacts-*
-          path: target/distrib/
-          merge-multiple: true
-      - name: Install dependencies
-        run: |
-          ${{ matrix.packages_install }}
-      - name: Build artifacts
-        run: |
-          # Actually do builds and make zips and whatnot
-          dist build ${{ needs.plan.outputs.tag-flag }} --print=linkage --output-format=json ${{ matrix.dist_args }} > dist-manifest.json
-          echo "dist ran successfully"
-      - id: cargo-dist
-        name: Post-build
-        # We force bash here just because github makes it really hard to get values up
-        # to "real" actions without writing to env-vars, and writing to env-vars has
-        # inconsistent syntax between shell and powershell.
-        shell: bash
-        run: |
-          # Parse out what we just built and upload it to scratch storage
-          echo "paths<<EOF" >> "$GITHUB_OUTPUT"
-          dist print-upload-files-from-manifest --manifest dist-manifest.json >> "$GITHUB_OUTPUT"
-          echo "EOF" >> "$GITHUB_OUTPUT"
-
-          cp dist-manifest.json "$BUILD_MANIFEST_NAME"
-      - name: "Upload artifacts"
-        uses: actions/upload-artifact@v4
-        with:
-          name: artifacts-build-local-${{ join(matrix.targets, '_') }}
-          path: |
-            ${{ steps.cargo-dist.outputs.paths }}
-            ${{ env.BUILD_MANIFEST_NAME }}
-
-  # Build and package all the platform-agnostic(ish) things
-  build-global-artifacts:
-    needs:
-      - plan
-      - build-local-artifacts
-    runs-on: "ubuntu-22.04"
-    env:
-      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-      BUILD_MANIFEST_NAME: target/distrib/global-dist-manifest.json
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          persist-credentials: false
-          submodules: recursive
-      - name: Install cached dist
-        uses: actions/download-artifact@v4
-        with:
-          name: cargo-dist-cache
-          path: ~/.cargo/bin/
-      - run: chmod +x ~/.cargo/bin/dist
-      # Get all the local artifacts for the global tasks to use (for e.g. checksums)
-      - name: Fetch local artifacts
-        uses: actions/download-artifact@v4
-        with:
-          pattern: artifacts-*
-          path: target/distrib/
-          merge-multiple: true
-      - id: cargo-dist
-        shell: bash
-        run: |
-          dist build ${{ needs.plan.outputs.tag-flag }} --output-format=json "--artifacts=global" > dist-manifest.json
-          echo "dist ran successfully"
-
-          # Parse out what we just built and upload it to scratch storage
-          echo "paths<<EOF" >> "$GITHUB_OUTPUT"
-          jq --raw-output ".upload_files[]" dist-manifest.json >> "$GITHUB_OUTPUT"
-          echo "EOF" >> "$GITHUB_OUTPUT"
-
-          cp dist-manifest.json "$BUILD_MANIFEST_NAME"
-      - name: "Upload artifacts"
-        uses: actions/upload-artifact@v4
-        with:
-          name: artifacts-build-global
-          path: |
-            ${{ steps.cargo-dist.outputs.paths }}
-            ${{ env.BUILD_MANIFEST_NAME }}
-  # Determines if we should publish/announce
-  host:
-    needs:
-      - plan
-      - build-local-artifacts
-      - build-global-artifacts
-    # Only run if we're "publishing", and only if local and global didn't fail (skipped is fine)
-    if: ${{ always() && needs.plan.outputs.publishing == 'true' && (needs.build-global-artifacts.result == 'skipped' || needs.build-global-artifacts.result == 'success') && (needs.build-local-artifacts.result == 'skipped' || needs.build-local-artifacts.result == 'success') }}
-    env:
-      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-    runs-on: "ubuntu-22.04"
-    outputs:
-      val: ${{ steps.host.outputs.manifest }}
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          persist-credentials: false
-          submodules: recursive
-      - name: Install cached dist
-        uses: actions/download-artifact@v4
-        with:
-          name: cargo-dist-cache
-          path: ~/.cargo/bin/
-      - run: chmod +x ~/.cargo/bin/dist
-      # Fetch artifacts from scratch-storage
-      - name: Fetch artifacts
-        uses: actions/download-artifact@v4
-        with:
-          pattern: artifacts-*
-          path: target/distrib/
-          merge-multiple: true
-      - id: host
-        shell: bash
-        run: |
-          dist host ${{ needs.plan.outputs.tag-flag }} --steps=upload --steps=release --output-format=json > dist-manifest.json
-          echo "artifacts uploaded and released successfully"
-          cat dist-manifest.json
-          echo "manifest=$(jq -c "." dist-manifest.json)" >> "$GITHUB_OUTPUT"
-      - name: "Upload dist-manifest.json"
-        uses: actions/upload-artifact@v4
-        with:
-          # Overwrite the previous copy
-          name: artifacts-dist-manifest
-          path: dist-manifest.json
-      # Create a GitHub Release while uploading all files to it
-      - name: "Download GitHub Artifacts"
-        uses: actions/download-artifact@v4
-        with:
-          pattern: artifacts-*
-          path: artifacts
-          merge-multiple: true
-      - name: Cleanup
-        run: |
-          # Remove the granular manifests
-          rm -f artifacts/*-dist-manifest.json
-      - name: Create GitHub Release
-        env:
-          PRERELEASE_FLAG: "${{ fromJson(steps.host.outputs.manifest).announcement_is_prerelease && '--prerelease' || '' }}"
-          ANNOUNCEMENT_TITLE: "${{ fromJson(steps.host.outputs.manifest).announcement_title }}"
-          ANNOUNCEMENT_BODY: "${{ fromJson(steps.host.outputs.manifest).announcement_github_body }}"
-          RELEASE_COMMIT: "${{ github.sha }}"
-        run: |
-          # Write and read notes from a file to avoid quoting breaking things
-          echo "$ANNOUNCEMENT_BODY" > $RUNNER_TEMP/notes.txt
-
-          gh release create "${{ needs.plan.outputs.tag }}" --target "$RELEASE_COMMIT" $PRERELEASE_FLAG --title "$ANNOUNCEMENT_TITLE" --notes-file "$RUNNER_TEMP/notes.txt" artifacts/*
-
-  announce:
-    needs:
-      - plan
-      - host
-    # use "always() && ..." to allow us to wait for all publish jobs while
-    # still allowing individual publish jobs to skip themselves (for prereleases).
-    # "host" however must run to completion, no skipping allowed!
-    if: ${{ always() && needs.host.result == 'success' }}
-    runs-on: "ubuntu-22.04"
-    env:
-      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          persist-credentials: false
-          submodules: recursive

.gitignore

@@ -6,15 +6,29 @@ target
 websites/dist
 bridge/
 /ids.txt
+rust_out
 
 # Copies
 *\ copy*
 
 # Ignored
 _*
+!__*.py
+/*.md
+/*.py
+/*.json
+/*.html
+!/btc-cycle-sim.html
+/research
+/filter_*
+/heatmaps*
+/oracle*
+/playground
+/*.txt
+/*.csv
 
 # Logs
-.log
+*.log*
 
 # Environment variables/configs
 .env
@@ -26,3 +40,14 @@ flamegraph.svg
 
 # AI
 .claude/settings*
+!CLAUDE.md
+
+# Expand
+expand.rs
+
+# Benchmarks
+[0-9]/
+/benches
+
+# AI
+.claude

.zed/settings.json

@@ -1,20 +0,0 @@
-{
-  "file_scan_exclusions": [
-    // default
-    "**/.git",
-    "**/.svn",
-    "**/.hg",
-    "**/.jj",
-    "**/CVS",
-    "**/.DS_Store",
-    "**/Thumbs.db",
-    "**/.classpath",
-    "**/.settings",
-    // custom
-    "**/lean-qr/*/index.mjs",
-    "uFuzzy.mjs",
-    "lightweight-charts.standalone.production.mjs",
-    "**/modern-screenshot/*/index.mjs",
-    "**/solidjs-signals/*/dist/prod.js"
-  ]
-}

Cargo.toml

@@ -4,11 +4,10 @@ members = ["crates/*"]
 package.description = "The Bitcoin Research Kit is a suite of tools designed to extract, compute and display data stored on a Bitcoin Core node"
 package.license = "MIT"
 package.edition = "2024"
-package.version = "0.0.109"
+package.version = "0.3.5"
 package.homepage = "https://bitcoinresearchkit.org"
 package.repository = "https://github.com/bitcoinresearchkit/brk"
 package.readme = "README.md"
-package.rust-version = "1.89"
 
 [profile.dev]
 lto = "thin"
@@ -23,71 +22,73 @@ panic = "abort"
 strip = true
 overflow-checks = false
 
+[profile.bloaty]
+debug = true
+lto = false
+strip = false
+inherits = "release"
+
 [profile.profiling]
 inherits = "release"
 debug = true
 
-[profile.dist]
-inherits = "release"
-
-[profile.clippy]
-inherits = "dev"
-lto = "off"
-codegen-units = 256
-opt-level = 0
-debug = false
-overflow-checks = false
-panic = "abort"
-debug-assertions = false
-
 [workspace.dependencies]
-allocative = { version = "0.3.4", features = ["parking_lot"] }
-axum = "0.8.4"
-bitcoin = { version = "0.32.7", features = ["serde"] }
-bitcoincore-rpc = "0.19.0"
-brk_bundler = { version = "0.0.109", path = "crates/brk_bundler" }
-brk_cli = { version = "0.0.109", path = "crates/brk_cli" }
-brk_computer = { version = "0.0.109", path = "crates/brk_computer" }
-brk_error = { version = "0.0.109", path = "crates/brk_error" }
-brk_fetcher = { version = "0.0.109", path = "crates/brk_fetcher" }
-brk_indexer = { version = "0.0.109", path = "crates/brk_indexer" }
-brk_interface = { version = "0.0.109", path = "crates/brk_interface" }
-brk_logger = { version = "0.0.109", path = "crates/brk_logger" }
-brk_mcp = { version = "0.0.109", path = "crates/brk_mcp" }
-brk_parser = { version = "0.0.109", path = "crates/brk_parser" }
-brk_server = { version = "0.0.109", path = "crates/brk_server" }
-brk_store = { version = "0.0.109", path = "crates/brk_store" }
-brk_structs = { version = "0.0.109", path = "crates/brk_structs" }
-byteview = "=0.6.1"
-derive_deref = "1.1.1"
-fjall = "2.11.2"
-jiff = "0.2.15"
-log = "0.4.28"
-minreq = { version = "2.14.1", features = ["https", "serde_json"] }
-parking_lot = "0.12.4"
-quick_cache = "0.6.16"
-rayon = "1.11.0"
-serde = "1.0.225"
+aide = { version = "0.16.0-alpha.4", features = ["axum-json", "axum-query"] }
+axum = { version = "0.8.9", default-features = false, features = ["http1", "json", "query", "tokio", "tracing"] }
+bitcoin = { version = "0.32.10", features = ["serde"] }
+brk_alloc = { version = "0.3.5", path = "crates/brk_alloc" }
+brk_bencher = { version = "0.3.5", path = "crates/brk_bencher" }
+brk_bindgen = { version = "0.3.5", path = "crates/brk_bindgen" }
+brk_cli = { version = "0.3.5", path = "crates/brk_cli" }
+brk_client = { version = "0.3.5", path = "crates/brk_client" }
+brk_cohort = { version = "0.3.5", path = "crates/brk_cohort" }
+brk_computer = { version = "0.3.5", path = "crates/brk_computer" }
+brk_error = { version = "0.3.5", path = "crates/brk_error" }
+brk_fetcher = { version = "0.3.5", path = "crates/brk_fetcher" }
+brk_indexer = { version = "0.3.5", path = "crates/brk_indexer" }
+brk_iterator = { version = "0.3.5", path = "crates/brk_iterator" }
+brk_logger = { version = "0.3.5", path = "crates/brk_logger" }
+brk_mempool = { version = "0.3.5", path = "crates/brk_mempool" }
+brk_oracle = { version = "0.3.5", path = "crates/brk_oracle" }
+brk_query = { version = "0.3.5", path = "crates/brk_query", features = ["tokio"] }
+brk_reader = { version = "0.3.5", path = "crates/brk_reader" }
+brk_rpc = { version = "0.3.5", path = "crates/brk_rpc" }
+brk_server = { version = "0.3.5", path = "crates/brk_server" }
+brk_store = { version = "0.3.5", path = "crates/brk_store" }
+brk_traversable = { version = "0.3.5", path = "crates/brk_traversable", features = ["pco", "derive"] }
+brk_traversable_derive = { version = "0.3.5", path = "crates/brk_traversable_derive" }
+brk_types = { version = "0.3.5", path = "crates/brk_types" }
+brk_website = { version = "0.3.5", path = "crates/brk_website" }
+byteview = "0.10.1"
+color-eyre = "0.6.5"
+corepc-jsonrpc = { package = "jsonrpc", version = "0.19.0", features = ["simple_http"], default-features = false }
+corepc-types = { version = "0.15.0", features = ["std"], default-features = false }
+derive_more = { version = "2.1.1", features = ["deref", "deref_mut"] }
+fjall = "3.1.5"
+indexmap = { version = "2.14.0", features = ["serde"] }
+jiff = { version = "0.2.29", features = ["perf-inline", "tz-system"], default-features = false }
+owo-colors = "4.3.0"
+parking_lot = "0.12.5"
+pco = "1.0.2"
+rayon = "1.12.0"
+rustc-hash = "2.1.2"
+schemars = { version = "1.2.1", features = ["indexmap2"] }
+serde = "1.0.228"
 serde_bytes = "0.11.19"
-serde_derive = "1.0.225"
-serde_json = { version = "1.0.145", features = ["float_roundtrip"] }
-sonic-rs = "0.5.4"
-tokio = { version = "1.47.1", features = ["rt-multi-thread"] }
-# vecdb = { path = "../seqdb/crates/vecdb", features = ["derive"]}
-vecdb = { version = "0.2.16", features = ["derive"]}
-zerocopy = "0.8.27"
-zerocopy-derive = "0.8.27"
+serde_derive = "1.0.228"
+serde_json = { version = "1.0.150", features = ["float_roundtrip", "preserve_order"] }
+smallvec = "1.15.2"
+tokio = { version = "1.52.3", features = ["rt-multi-thread"] }
+tower-http = { version = "0.7.0", features = ["catch-panic", "compression-br", "compression-gzip", "compression-zstd", "cors", "normalize-path", "timeout", "trace"] }
+tower-layer = "0.3"
+tracing = { version = "0.1", default-features = false, features = ["std"] }
+ureq = { version = "3.3.0", features = ["json"] }
+vecdb = { version = "0.10.3", features = ["derive", "serde_json", "pco", "schemars"] }
+# vecdb = { path = "../anydb/crates/vecdb", features = ["derive", "serde_json", "pco", "schemars"] }
 
 [workspace.metadata.release]
 shared-version = true
 tag-name = "v{{version}}"
 pre-release-commit-message = "release: v{{version}}"
 tag-message = "release: v{{version}}"
-
-[workspace.metadata.dist]
-cargo-dist-version = "0.30.0"
-ci = "github"
-allow-dirty = ["ci"]
-installers = []
-targets = ["aarch64-apple-darwin", "aarch64-unknown-linux-gnu", "x86_64-unknown-linux-gnu"]
-rust-toolchain-version = "1.89"
+allow-branch = ["main", "next"]

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2025 bitcoinresearchkit, kibo.money, satonomics
+Copyright (c) 2025 Bitcoin Research Kit
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

crates/blk/.gitignore

@@ -0,0 +1,2 @@
+*.md
+!README.md

crates/blk/Cargo.toml

@@ -0,0 +1,21 @@
+[package]
+name = "blk"
+description = "A CLI to inspect Bitcoin Core blocks"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+homepage.workspace = true
+repository.workspace = true
+
+[dependencies]
+bitcoin = { workspace = true }
+brk_error = { workspace = true }
+brk_reader = { workspace = true }
+brk_rpc = { workspace = true }
+brk_types = { workspace = true }
+owo-colors = { workspace = true, features = ["supports-colors"] }
+serde_json = { workspace = true }
+
+[[bin]]
+name = "blk"
+path = "src/main.rs"

crates/blk/README.md

@@ -0,0 +1,27 @@
+# blk
+
+A CLI to inspect Bitcoin Core blocks.
+
+Reads `blk*.dat` files directly via [`brk_reader`](../brk_reader) and resolves
+the chain tip / heights via the Bitcoin Core RPC. Output is shell-friendly:
+bare values, NDJSON, pretty JSON, or TSV.
+
+## Install
+
+```sh
+cargo install --path crates/blk
+```
+
+## Quick start
+
+```sh
+blk 800000 hash                        # bare hash
+blk 800000 height hash time            # one compact JSON line
+blk 800000 tx.0.vout.0.value           # coinbase output 0 sats
+blk 0..2 hash tx.0.txid                # 3 NDJSON lines
+blk tip tx.0                           # whole coinbase tx as JSON
+```
+
+## Reference
+
+Run `blk --help` for the full field/selector/option reference.

crates/blk/src/args.rs

@@ -0,0 +1,132 @@
+use std::{collections::HashSet, path::PathBuf};
+
+use brk_error::{Error, Result};
+use brk_rpc::{Auth, Client};
+
+use crate::path::Path;
+
+pub struct Args {
+    pub selector: String,
+    pub paths: Vec<Path>,
+    pub pretty: bool,
+    pub compact: bool,
+    bitcoindir: Option<PathBuf>,
+    blocksdir: Option<PathBuf>,
+    rpcconnect: Option<String>,
+    rpcport: Option<u16>,
+    rpccookiefile: Option<PathBuf>,
+    rpcuser: Option<String>,
+    rpcpassword: Option<String>,
+}
+
+impl Args {
+    pub fn parse(raw: Vec<String>) -> Result<Self> {
+        let mut pretty = false;
+        let mut compact = false;
+        let mut bitcoindir = None;
+        let mut blocksdir = None;
+        let mut rpcconnect = None;
+        let mut rpcport = None;
+        let mut rpccookiefile = None;
+        let mut rpcuser = None;
+        let mut rpcpassword = None;
+        let mut positional: Vec<String> = Vec::new();
+        let mut iter = raw.into_iter();
+        while let Some(a) = iter.next() {
+            if a == "-p" || a == "--pretty" {
+                pretty = true;
+                continue;
+            }
+            if a == "-c" || a == "--compact" {
+                compact = true;
+                continue;
+            }
+            if let Some(rest) = a.strip_prefix("--") {
+                let (key, value) = match rest.split_once('=') {
+                    Some((k, v)) => (k.to_string(), v.to_string()),
+                    None => (
+                        rest.to_string(),
+                        iter.next()
+                            .ok_or_else(|| Error::Parse(format!("--{rest} requires a value")))?,
+                    ),
+                };
+                match key.as_str() {
+                    "bitcoindir" => bitcoindir = Some(PathBuf::from(value)),
+                    "blocksdir" => blocksdir = Some(PathBuf::from(value)),
+                    "rpcconnect" => rpcconnect = Some(value),
+                    "rpcport" => {
+                        rpcport = Some(value.parse().map_err(|_| {
+                            Error::Parse(format!("--rpcport: '{value}' is not a valid port"))
+                        })?);
+                    }
+                    "rpccookiefile" => rpccookiefile = Some(PathBuf::from(value)),
+                    "rpcuser" => rpcuser = Some(value),
+                    "rpcpassword" => rpcpassword = Some(value),
+                    other => return Err(Error::Parse(format!("unknown flag --{other}"))),
+                }
+                continue;
+            }
+            if a.starts_with('-') {
+                return Err(Error::Parse(format!("unknown flag {a}")));
+            }
+            positional.push(a);
+        }
+
+        let mut iter = positional.into_iter();
+        let selector = iter
+            .next()
+            .ok_or_else(|| Error::Parse("missing selector".into()))?;
+        let paths: Vec<Path> = iter.map(|f| Path::parse(&f)).collect::<Result<_>>()?;
+        let mut seen = HashSet::with_capacity(paths.len());
+        for p in &paths {
+            if !seen.insert(p.raw.as_str()) {
+                return Err(Error::Parse(format!("duplicate field '{}'", p.raw)));
+            }
+        }
+        Ok(Self {
+            selector,
+            paths,
+            pretty,
+            compact,
+            bitcoindir,
+            blocksdir,
+            rpcconnect,
+            rpcport,
+            rpccookiefile,
+            rpcuser,
+            rpcpassword,
+        })
+    }
+
+    pub fn bitcoin_dir(&self) -> PathBuf {
+        self.bitcoindir
+            .clone()
+            .unwrap_or_else(Client::default_bitcoin_path)
+    }
+
+    pub fn blocks_dir(&self) -> PathBuf {
+        self.blocksdir
+            .clone()
+            .unwrap_or_else(|| self.bitcoin_dir().join("blocks"))
+    }
+
+    pub fn rpc(&self) -> Result<Client> {
+        let host = self.rpcconnect.as_deref().unwrap_or("localhost");
+        let port = self.rpcport.unwrap_or(8332);
+        let url = format!("http://{host}:{port}");
+        let cookie = self
+            .rpccookiefile
+            .clone()
+            .unwrap_or_else(|| self.bitcoin_dir().join(".cookie"));
+        let auth = if cookie.is_file() {
+            Auth::CookieFile(cookie)
+        } else if let (Some(u), Some(p)) = (self.rpcuser.as_deref(), self.rpcpassword.as_deref()) {
+            Auth::UserPass(u.to_string(), p.to_string())
+        } else {
+            return Err(Error::Parse(
+                "no RPC auth: cookie file missing and --rpcuser/--rpcpassword not set".into(),
+            ));
+        };
+        Client::new(&url, auth)
+    }
+}

crates/blk/src/fields.rs

@@ -0,0 +1,407 @@
+use std::cell::OnceCell;
+
+use bitcoin::{
+    Address, Block, Network, ScriptBuf, Transaction, TxIn, TxOut, consensus::encode::serialize_hex,
+    hex::DisplayHex,
+};
+use brk_error::{Error, Result};
+use brk_types::ReadBlock;
+use serde_json::{Map, Value, json};
+
+use crate::path::{Path, Step};
+
+// `hex` is intentionally absent: matches `bitcoin-cli getblock <hash> 2`
+// and keeps NDJSON dumps tractable. Still reachable explicitly via `blk N hex`.
+const BLOCK_FIELDS: &[&str] = &[
+    "height",
+    "hash",
+    "version",
+    "version_hex",
+    "merkle",
+    "time",
+    "nonce",
+    "bits",
+    "difficulty",
+    "prev",
+    "txs",
+    "n_inputs",
+    "n_outputs",
+    "witness_txs",
+    "size",
+    "strippedsize",
+    "weight",
+    "subsidy",
+    "coinbase",
+    "coinbase_hex",
+    "header_hex",
+    "tx",
+];
+
+const TX_FIELDS: &[&str] = &[
+    "txid",
+    "wtxid",
+    "version",
+    "locktime",
+    "size",
+    "base_size",
+    "vsize",
+    "weight",
+    "inputs",
+    "outputs",
+    "is_coinbase",
+    "has_witness",
+    "is_rbf",
+    "total_out",
+    "hex",
+    "vin",
+    "vout",
+];
+
+const VIN_FIELDS: &[&str] = &[
+    "prev_txid",
+    "prev_vout",
+    "sequence",
+    "script_sig",
+    "script_sig_asm",
+    "witness",
+    "has_witness",
+    "is_rbf",
+    "coinbase",
+];
+
+const VOUT_FIELDS: &[&str] = &[
+    "value",
+    "script_pubkey",
+    "script_pubkey_asm",
+    "type",
+    "address",
+];
+
+pub struct Ctx<'a> {
+    block: &'a ReadBlock,
+    network: Network,
+    size_weight: OnceCell<(usize, usize)>,
+}
+
+impl<'a> Ctx<'a> {
+    pub fn new(block: &'a ReadBlock, network: Network) -> Self {
+        Self {
+            block,
+            network,
+            size_weight: OnceCell::new(),
+        }
+    }
+
+    pub fn resolve(&self, path: &Path) -> Result<Value> {
+        let (step, rest) = pop(&path.steps)?;
+        self.block_field(&step.name, step.index, rest)
+    }
+
+    pub fn resolve_str(&self, path: &Path) -> Result<String> {
+        Ok(match self.resolve(path)? {
+            Value::String(s) => s,
+            other => other.to_string(),
+        })
+    }
+
+    pub fn full(&self) -> Value {
+        let mut obj = Map::with_capacity(BLOCK_FIELDS.len());
+        for &name in BLOCK_FIELDS {
+            obj.insert(
+                name.into(),
+                self.block_field(name, None, &[]).expect("known block field"),
+            );
+        }
+        Value::Object(obj)
+    }
+
+    fn size_and_weight(&self) -> (usize, usize) {
+        *self
+            .size_weight
+            .get_or_init(|| self.block.total_size_and_weight())
+    }
+
+    fn block_field(&self, name: &str, index: Option<usize>, rest: &[Step]) -> Result<Value> {
+        let b = self.block;
+        let raw: &Block = b;
+        let scalar = |v| scalar_leaf(v, name, index, rest);
+        match name {
+            "height" => scalar(json!(*b.height())),
+            "hash" => scalar(json!(b.hash().to_string())),
+            "time" => scalar(json!(b.header.time)),
+            "version" => scalar(json!(b.header.version.to_consensus())),
+            "version_hex" => scalar(json!(format!(
+                "{:08x}",
+                b.header.version.to_consensus() as u32
+            ))),
+            "bits" => scalar(json!(format!("{:08x}", b.header.bits.to_consensus()))),
+            "nonce" => scalar(json!(b.header.nonce)),
+            "prev" => scalar(json!(b.header.prev_blockhash.to_string())),
+            "merkle" => scalar(json!(b.header.merkle_root.to_string())),
+            "difficulty" => scalar(json!(b.header.difficulty_float())),
+            "txs" => scalar(json!(b.txdata.len())),
+            "n_inputs" => scalar(json!(
+                b.txdata.iter().map(|tx| tx.input.len()).sum::<usize>()
+            )),
+            "n_outputs" => scalar(json!(
+                b.txdata.iter().map(|tx| tx.output.len()).sum::<usize>()
+            )),
+            "witness_txs" => scalar(json!(
+                b.txdata.iter().filter(|tx| tx_has_witness(tx)).count()
+            )),
+            "size" => scalar(json!(self.size_and_weight().0)),
+            "weight" => scalar(json!(self.size_and_weight().1)),
+            "strippedsize" => {
+                let (size, weight) = self.size_and_weight();
+                scalar(json!((weight - size) / 3))
+            }
+            "subsidy" => scalar(json!(subsidy_sats(*b.height()))),
+            "header_hex" => scalar(json!(serialize_hex(&b.header))),
+            "hex" => scalar(json!(serialize_hex(raw))),
+            "coinbase" => scalar(json!(b.coinbase_tag().as_str())),
+            "coinbase_hex" => {
+                debug_assert!(
+                    !b.txdata.is_empty() && !b.txdata[0].input.is_empty(),
+                    "consensus-valid block has a coinbase tx with at least one input"
+                );
+                scalar(json!(b.txdata[0].input[0].script_sig.to_hex_string()))
+            }
+            "tx" => pick(&b.txdata, name, index, |i, tx| {
+                self.resolve_tx(tx, i == 0, rest)
+            }),
+            other => Err(unknown("block", other)),
+        }
+    }
+
+    fn resolve_tx(&self, tx: &Transaction, is_coinbase: bool, steps: &[Step]) -> Result<Value> {
+        if steps.is_empty() {
+            let mut obj = Map::with_capacity(TX_FIELDS.len());
+            for &name in TX_FIELDS {
+                obj.insert(
+                    name.into(),
+                    self.tx_field(tx, is_coinbase, name, None, &[])
+                        .expect("known tx field"),
+                );
+            }
+            return Ok(Value::Object(obj));
+        }
+        let (step, rest) = pop(steps)?;
+        self.tx_field(tx, is_coinbase, &step.name, step.index, rest)
+    }
+
+    fn tx_field(
+        &self,
+        tx: &Transaction,
+        is_coinbase: bool,
+        name: &str,
+        index: Option<usize>,
+        rest: &[Step],
+    ) -> Result<Value> {
+        let scalar = |v| scalar_leaf(v, name, index, rest);
+        match name {
+            "txid" => scalar(json!(tx.compute_txid().to_string())),
+            "wtxid" => scalar(json!(tx.compute_wtxid().to_string())),
+            "version" => scalar(json!(tx.version.0)),
+            "locktime" => scalar(json!(tx.lock_time.to_consensus_u32())),
+            "size" => scalar(json!(tx.total_size())),
+            "base_size" => scalar(json!(tx.base_size())),
+            "vsize" => scalar(json!(tx.vsize())),
+            "weight" => scalar(json!(tx.weight().to_wu())),
+            "inputs" => scalar(json!(tx.input.len())),
+            "outputs" => scalar(json!(tx.output.len())),
+            "is_coinbase" => scalar(json!(is_coinbase)),
+            "has_witness" => scalar(json!(tx_has_witness(tx))),
+            "is_rbf" => scalar(json!(tx_is_rbf(tx))),
+            "total_out" => scalar(json!(tx_total_out(tx))),
+            "hex" => scalar(json!(serialize_hex(tx))),
+            "vin" => pick(&tx.input, name, index, |j, vin| {
+                resolve_vin(vin, is_coinbase && j == 0, rest)
+            }),
+            "vout" => pick(&tx.output, name, index, |_, vout| {
+                self.resolve_vout(vout, rest)
+            }),
+            other => Err(unknown("tx", other)),
+        }
+    }
+
+    fn resolve_vout(&self, vout: &TxOut, steps: &[Step]) -> Result<Value> {
+        if steps.is_empty() {
+            let mut obj = Map::with_capacity(VOUT_FIELDS.len());
+            for &name in VOUT_FIELDS {
+                obj.insert(
+                    name.into(),
+                    self.vout_field(vout, name, None, &[])
+                        .expect("known vout field"),
+                );
+            }
+            return Ok(Value::Object(obj));
+        }
+        let (step, rest) = pop(steps)?;
+        self.vout_field(vout, &step.name, step.index, rest)
+    }
+
+    fn vout_field(
+        &self,
+        vout: &TxOut,
+        name: &str,
+        index: Option<usize>,
+        rest: &[Step],
+    ) -> Result<Value> {
+        let scalar = |v| scalar_leaf(v, name, index, rest);
+        match name {
+            "value" => scalar(json!(vout.value.to_sat())),
+            "script_pubkey" => scalar(json!(vout.script_pubkey.to_hex_string())),
+            "script_pubkey_asm" => scalar(json!(vout.script_pubkey.to_asm_string())),
+            "type" => scalar(json!(script_type(&vout.script_pubkey))),
+            "address" => scalar(self.address_value(&vout.script_pubkey)),
+            other => Err(unknown("vout", other)),
+        }
+    }
+
+    fn address_value(&self, s: &ScriptBuf) -> Value {
+        Address::from_script(s, self.network)
+            .map(|a| Value::String(a.to_string()))
+            .unwrap_or(Value::Null)
+    }
+}
+
+fn resolve_vin(vin: &TxIn, is_coinbase: bool, steps: &[Step]) -> Result<Value> {
+    if steps.is_empty() {
+        let mut obj = Map::with_capacity(VIN_FIELDS.len());
+        for &name in VIN_FIELDS {
+            obj.insert(
+                name.into(),
+                vin_field(vin, is_coinbase, name, None, &[]).expect("known vin field"),
+            );
+        }
+        return Ok(Value::Object(obj));
+    }
+    let (step, rest) = pop(steps)?;
+    vin_field(vin, is_coinbase, &step.name, step.index, rest)
+}
+
+fn vin_field(
+    vin: &TxIn,
+    is_coinbase: bool,
+    name: &str,
+    index: Option<usize>,
+    rest: &[Step],
+) -> Result<Value> {
+    let scalar = |v| scalar_leaf(v, name, index, rest);
+    match name {
+        "prev_txid" => scalar(json!(vin.previous_output.txid.to_string())),
+        "prev_vout" => scalar(json!(vin.previous_output.vout)),
+        "sequence" => scalar(json!(vin.sequence.0)),
+        "script_sig" => scalar(json!(vin.script_sig.to_hex_string())),
+        "script_sig_asm" => scalar(json!(vin.script_sig.to_asm_string())),
+        "witness" => {
+            if !rest.is_empty() {
+                return Err(Error::Parse(
+                    "'witness' element has no fields to drill into".into(),
+                ));
+            }
+            let items: Vec<String> = vin
+                .witness
+                .iter()
+                .map(|w| w.to_lower_hex_string())
+                .collect();
+            pick(&items, name, index, |_, hex| Ok(Value::String(hex.clone())))
+        }
+        "has_witness" => scalar(json!(!vin.witness.is_empty())),
+        "is_rbf" => scalar(json!(vin.sequence.is_rbf())),
+        "coinbase" => scalar(json!(is_coinbase)),
+        other => Err(unknown("vin", other)),
+    }
+}
+
+fn pick<T>(
+    items: &[T],
+    name: &str,
+    index: Option<usize>,
+    mut resolve: impl FnMut(usize, &T) -> Result<Value>,
+) -> Result<Value> {
+    match index {
+        Some(i) => {
+            let item = items
+                .get(i)
+                .ok_or_else(|| out_of_range(name, i, items.len()))?;
+            resolve(i, item)
+        }
+        None => Ok(Value::Array(
+            items
+                .iter()
+                .enumerate()
+                .map(|(i, item)| resolve(i, item))
+                .collect::<Result<_>>()?,
+        )),
+    }
+}
+
+fn pop(steps: &[Step]) -> Result<(&Step, &[Step])> {
+    steps
+        .split_first()
+        .ok_or_else(|| Error::Parse("empty path segment".into()))
+}
+
+fn scalar_leaf(v: Value, name: &str, index: Option<usize>, rest: &[Step]) -> Result<Value> {
+    if index.is_some() {
+        return Err(Error::Parse(format!("'{name}' is not an array")));
+    }
+    if !rest.is_empty() {
+        return Err(Error::Parse(format!(
+            "'{name}' has no fields to drill into"
+        )));
+    }
+    Ok(v)
+}
+
+fn out_of_range(name: &str, i: usize, len: usize) -> Error {
+    Error::Parse(format!("{name}.{i} out of range (len {len})"))
+}
+
+fn unknown(level: &str, name: &str) -> Error {
+    Error::Parse(format!(
+        "unknown {level} field '{name}' (run `blk --help` for the list)"
+    ))
+}
+
+fn tx_has_witness(tx: &Transaction) -> bool {
+    tx.input.iter().any(|i| !i.witness.is_empty())
+}
+
+fn tx_is_rbf(tx: &Transaction) -> bool {
+    tx.input.iter().any(|i| i.sequence.is_rbf())
+}
+
+fn tx_total_out(tx: &Transaction) -> u64 {
+    tx.output.iter().map(|o| o.value.to_sat()).sum()
+}
+
+fn subsidy_sats(height: u32) -> u64 {
+    let halvings = height / 210_000;
+    if halvings >= 64 {
+        0
+    } else {
+        (50 * 100_000_000u64) >> halvings
+    }
+}
+
+fn script_type(s: &ScriptBuf) -> &'static str {
+    if s.is_p2pkh() {
+        "p2pkh"
+    } else if s.is_p2sh() {
+        "p2sh"
+    } else if s.is_p2wpkh() {
+        "p2wpkh"
+    } else if s.is_p2wsh() {
+        "p2wsh"
+    } else if s.is_p2tr() {
+        "p2tr"
+    } else if s.is_op_return() {
+        "op_return"
+    } else if s.is_p2pk() {
+        "p2pk"
+    } else {
+        "unknown"
+    }
+}

crates/blk/src/formatter.rs

@@ -0,0 +1,61 @@
+use brk_error::Result;
+use serde_json::{Map, Value};
+
+use crate::{fields::Ctx, mode::Mode, path::Path};
+
+pub struct Formatter {
+    mode: Mode,
+    fields: Vec<Path>,
+}
+
+impl Formatter {
+    pub fn new(mode: Mode, fields: Vec<Path>) -> Self {
+        Self { mode, fields }
+    }
+
+    pub fn format(&self, ctx: &Ctx) -> Result<String> {
+        match self.mode {
+            Mode::Bare => self.bare(ctx, false),
+            Mode::Tsv => self.tsv(ctx),
+            Mode::Json => Ok(serde_json::to_string(&self.object(ctx)?)?),
+            Mode::Pretty if self.fields.len() == 1 => self.bare(ctx, true),
+            Mode::Pretty => Ok(serde_json::to_string_pretty(&self.object(ctx)?)?),
+        }
+    }
+
+    fn bare(&self, ctx: &Ctx, pretty: bool) -> Result<String> {
+        Ok(match ctx.resolve(&self.fields[0])? {
+            Value::String(s) => s,
+            other if pretty => serde_json::to_string_pretty(&other)?,
+            other => other.to_string(),
+        })
+    }
+
+    fn tsv(&self, ctx: &Ctx) -> Result<String> {
+        let mut row = String::new();
+        for (i, path) in self.fields.iter().enumerate() {
+            if i > 0 {
+                row.push('\t');
+            }
+            for c in ctx.resolve_str(path)?.chars() {
+                row.push(if matches!(c, '\t' | '\n' | '\r') {
+                    ' '
+                } else {
+                    c
+                });
+            }
+        }
+        Ok(row)
+    }
+
+    fn object(&self, ctx: &Ctx) -> Result<Value> {
+        if self.fields.is_empty() {
+            return Ok(ctx.full());
+        }
+        let mut obj = Map::with_capacity(self.fields.len());
+        for path in &self.fields {
+            obj.insert(path.raw.clone(), ctx.resolve(path)?);
+        }
+        Ok(Value::Object(obj))
+    }
+}

crates/blk/src/main.rs

@@ -0,0 +1,58 @@
+mod args;
+mod fields;
+mod formatter;
+mod mode;
+mod path;
+mod selector;
+mod usage;
+
+use std::process::ExitCode;
+
+use brk_error::Result;
+use brk_reader::Reader;
+
+use args::Args;
+use fields::Ctx;
+use formatter::Formatter;
+use mode::Mode;
+use selector::Selector;
+
+fn main() -> ExitCode {
+    match run() {
+        Ok(()) => ExitCode::SUCCESS,
+        Err(e) => {
+            eprintln!("blk: {e}");
+            ExitCode::from(1)
+        }
+    }
+}
+
+fn run() -> Result<()> {
+    let raw: Vec<String> = std::env::args().skip(1).collect();
+    if raw.is_empty() || raw.iter().any(|a| matches!(a.as_str(), "-h" | "--help")) {
+        usage::print();
+        return Ok(());
+    }
+    let args = Args::parse(raw)?;
+
+    let client = args.rpc()?;
+    let (start, end) = Selector::parse(&args.selector, &client)?;
+    let network = client.get_network()?;
+
+    let mode = Mode::pick(args.pretty, args.compact, args.paths.len())?;
+    let reader = Reader::new(args.blocks_dir(), &client);
+    let formatter = Formatter::new(mode, args.paths);
+    let parser_threads = (std::thread::available_parallelism()
+        .map(|n| n.get())
+        .unwrap_or(2)
+        / 2)
+        .max(1);
+    for block in reader.range_with(start, end, parser_threads)?.iter() {
+        let block = block?;
+        let line = formatter.format(&Ctx::new(&block, network))?;
+        if !line.is_empty() {
+            println!("{line}");
+        }
+    }
+    Ok(())
+}

crates/blk/src/mode.rs

@@ -0,0 +1,35 @@
+use brk_error::{Error, Result};
+
+#[derive(Clone, Copy)]
+pub enum Mode {
+    Bare,
+    Tsv,
+    Json,
+    Pretty,
+}
+
+impl Mode {
+    pub fn pick(pretty: bool, compact: bool, n_fields: usize) -> Result<Self> {
+        if pretty && compact {
+            return Err(Error::Parse(
+                "--pretty and --compact are mutually exclusive".into(),
+            ));
+        }
+        if compact && n_fields == 0 {
+            return Err(Error::Parse(
+                "--compact requires at least one field".into(),
+            ));
+        }
+        Ok(if pretty {
+            Self::Pretty
+        } else if n_fields == 0 {
+            Self::Json
+        } else if n_fields == 1 {
+            Self::Bare
+        } else if compact {
+            Self::Tsv
+        } else {
+            Self::Json
+        })
+    }
+}

crates/blk/src/path.rs

@@ -0,0 +1,40 @@
+use brk_error::{Error, Result};
+
+pub struct Step {
+    pub name: String,
+    pub index: Option<usize>,
+}
+
+pub struct Path {
+    pub raw: String,
+    pub steps: Vec<Step>,
+}
+
+impl Path {
+    pub fn parse(s: &str) -> Result<Self> {
+        let parts: Vec<&str> = s.split('.').collect();
+        let mut steps = Vec::new();
+        let mut i = 0;
+        while i < parts.len() {
+            let name = parts[i];
+            if name.is_empty() {
+                return Err(Error::Parse(format!("bad path '{s}': empty segment")));
+            }
+            if name.parse::<usize>().is_ok() {
+                return Err(Error::Parse(format!(
+                    "bad path '{s}': '{name}' must follow a field name"
+                )));
+            }
+            let index = parts.get(i + 1).and_then(|p| p.parse::<usize>().ok());
+            steps.push(Step {
+                name: name.to_string(),
+                index,
+            });
+            i += if index.is_some() { 2 } else { 1 };
+        }
+        Ok(Self {
+            raw: s.to_string(),
+            steps,
+        })
+    }
+}

crates/blk/src/selector.rs

@@ -0,0 +1,44 @@
+use brk_error::{Error, Result};
+use brk_rpc::Client;
+use brk_types::{CheckedSub, Height};
+
+pub struct Selector;
+
+impl Selector {
+    pub fn parse(s: &str, client: &Client) -> Result<(Height, Height)> {
+        let (a, b) = s.split_once("..").unwrap_or((s, s));
+        let needs_tip = |p: &str| p == "tip" || p.starts_with("tip-");
+        let tip = if needs_tip(a) || needs_tip(b) {
+            Some(client.get_last_height()?)
+        } else {
+            None
+        };
+        let start = Self::endpoint(a, tip)?;
+        let end = Self::endpoint(b, tip)?;
+        if end < start {
+            return Err(Error::Parse(format!(
+                "range end {end} before start {start}"
+            )));
+        }
+        Ok((start, end))
+    }
+
+    fn endpoint(s: &str, tip: Option<Height>) -> Result<Height> {
+        if s == "tip" {
+            return Ok(tip.expect("tip pre-resolved when input contains 'tip'"));
+        }
+        if let Some(rest) = s.strip_prefix("tip-") {
+            let n: u32 = rest
+                .parse()
+                .map_err(|_| Error::Parse(format!("bad tip offset: {s}")))?;
+            let tip = tip.expect("tip pre-resolved when input contains 'tip'");
+            return tip
+                .checked_sub(n)
+                .ok_or_else(|| Error::Parse(format!("tip-{n} underflows genesis")));
+        }
+        let n: u32 = s
+            .parse()
+            .map_err(|_| Error::Parse(format!("bad height: {s}")))?;
+        Ok(Height::new(n))
+    }
+}

crates/blk/src/usage.rs

@@ -0,0 +1,190 @@
+use owo_colors::{OwoColorize, Stream};
+
+const SEL_W: usize = 5; // longest selector token: "tip-N"
+const LABEL_W: usize = 28; // longest label across OUTPUT/OPTIONS/EXAMPLES (= example cmd "blk 800000 tx.0.vout.0.value")
+const FLAG_W: usize = 15; // longest flag: "--rpccookiefile"
+const PH_W: usize = LABEL_W - FLAG_W - 1; // placeholder column width so flag+ph total = LABEL_W
+const GAP: usize = 4;
+
+pub fn print() {
+    println!("{} - inspect a Bitcoin Core block", bold("blk"));
+    println!();
+
+    section("USAGE");
+    println!(
+        "    blk {} [{} ...] [OPTIONS]",
+        dim("<selector>"),
+        dim("<field>")
+    );
+    println!(
+        "    {}",
+        dim("no fields = full block as JSON (analog of `bitcoin-cli getblock <hash> 2`)")
+    );
+    println!();
+
+    section("SELECTOR");
+    sel("<n>", "single height (e.g. 800000)");
+    sel("tip", "current chain tip");
+    sel("tip-N", "tip minus N");
+    sel("a..b", "inclusive range, endpoints can be height/tip/tip-N");
+    println!();
+
+    section("FIELDS");
+    println!(
+        "    {}",
+        dim("dotted paths drill into nested data, omit an index for arrays")
+    );
+    println!();
+    group("block");
+    fields(&[
+        "height, hash, time, version, version_hex, bits, nonce,",
+        "prev, merkle, difficulty, txs, n_inputs, n_outputs,",
+        "witness_txs, size, strippedsize, weight, subsidy,",
+        "coinbase, coinbase_hex, header_hex, hex",
+    ]);
+    println!();
+    group_note("tx.i", "omit i for all txs");
+    fields(&[
+        "txid, wtxid, version, locktime, size, base_size, vsize,",
+        "weight, inputs, outputs, is_coinbase, has_witness, is_rbf,",
+        "total_out, hex",
+    ]);
+    println!();
+    group_note("tx.i.vin.j", "omit j for all inputs");
+    fields(&[
+        "prev_txid, prev_vout, sequence, script_sig, script_sig_asm,",
+        "witness, has_witness, is_rbf, coinbase",
+    ]);
+    println!();
+    group_note("tx.i.vout.j", "omit j for all outputs");
+    fields(&["value, script_pubkey, script_pubkey_asm, type, address"]);
+    println!();
+    println!(
+        "    {}",
+        dim("Naked tx / tx.i / vin / vout returns the whole sub-object as JSON.")
+    );
+    println!();
+
+    section("OUTPUT");
+    out("no fields", "full block JSON object, one per line (NDJSON)");
+    out("1 field", "bare value, one per line");
+    out("2+ fields", "JSON object, one per line (NDJSON)");
+    out("-p, --pretty", "pretty JSON object instead");
+    out(
+        "-c, --compact",
+        "tab-separated values, no field names (TSV)",
+    );
+    println!();
+
+    section("OPTIONS");
+    opt(
+        "--bitcoindir",
+        "<PATH>",
+        "Bitcoin directory",
+        Some("[OS default]"),
+    );
+    opt(
+        "--blocksdir",
+        "<PATH>",
+        "Blocks directory",
+        Some("[<bitcoindir>/blocks]"),
+    );
+    opt("--rpcconnect", "<IP>", "RPC host", Some("[localhost]"));
+    opt("--rpcport", "<PORT>", "RPC port", Some("[8332]"));
+    opt(
+        "--rpccookiefile",
+        "<PATH>",
+        "RPC cookie file",
+        Some("[<bitcoindir>/.cookie]"),
+    );
+    opt("--rpcuser", "<USERNAME>", "RPC username", None);
+    opt("--rpcpassword", "<PASSWORD>", "RPC password", None);
+    println!();
+
+    section("EXAMPLES");
+    ex("blk 800000", "full block as JSON");
+    ex("blk 800000 hash", "bare hash");
+    ex("blk 800000 height hash time", "one compact JSON line");
+    ex("blk 800000 tx.0.txid", "coinbase txid");
+    ex("blk 800000 tx.txid", "all txids in block (array)");
+    ex("blk 800000 tx.0.vout.0.value", "coinbase output 0 sats");
+    ex("blk 800000 tx.0.vout.value", "all output sats for tx 0");
+    ex("blk 800000 tx.vout.value", "array of arrays (per tx)");
+    ex("blk 0..2 hash tx.0.txid", "3 NDJSON lines");
+    ex("blk tip tx.0", "whole coinbase tx as JSON");
+}
+
+fn section(name: &str) {
+    println!("{}", bold(&format!("{name}:")));
+}
+
+fn group(name: &str) {
+    println!("  {}", bold(&format!("{name}:")));
+}
+
+fn group_note(name: &str, note: &str) {
+    println!(
+        "  {}  {}",
+        bold(&format!("{name}:")),
+        dim(&format!("({note})"))
+    );
+}
+
+fn fields(lines: &[&str]) {
+    for line in lines {
+        println!("    {line}");
+    }
+}
+
+fn pad(s: &str, width: usize) -> String {
+    " ".repeat(width.saturating_sub(s.len()))
+}
+
+fn sel(token: &str, desc: &str) {
+    println!(
+        "    {}{}{}{desc}",
+        dim(token),
+        pad(token, SEL_W),
+        " ".repeat(GAP),
+    );
+}
+
+fn out(label: &str, desc: &str) {
+    println!(
+        "    {label}{}{}{desc}",
+        pad(label, LABEL_W),
+        " ".repeat(GAP)
+    );
+}
+
+fn opt(flag: &str, ph: &str, desc: &str, default: Option<&str>) {
+    let head = format!(
+        "    {flag}{} {}{}{}",
+        pad(flag, FLAG_W),
+        dim(ph),
+        pad(ph, PH_W),
+        " ".repeat(GAP),
+    );
+    match default {
+        Some(d) => println!("{head}{desc} {}", dim(d)),
+        None => println!("{head}{desc}"),
+    }
+}
+
+fn ex(cmd: &str, note: &str) {
+    println!(
+        "    {cmd}{}{}{}",
+        pad(cmd, LABEL_W),
+        " ".repeat(GAP),
+        dim(&format!("# {note}"))
+    );
+}
+
+fn bold(s: &str) -> String {
+    s.if_supports_color(Stream::Stdout, |t| t.bold()).to_string()
+}
+
+fn dim(s: &str) -> String {
+    s.if_supports_color(Stream::Stdout, |t| t.bright_black())
+        .to_string()
+}

crates/brk/Cargo.toml

@@ -6,51 +6,69 @@ homepage.workspace = true
 repository.workspace = true
 edition.workspace = true
 version.workspace = true
-rust-version.workspace = true
-build = "build.rs"
 
 [features]
 full = [
-    "bundler",
+    "bencher",
+    "bindgen",
+    "client",
     "computer",
     "error",
     "fetcher",
+    "cohort",
     "indexer",
-    "interface",
+    "iterator",
     "logger",
-    "mcp",
-    "parser",
+    "mempool",
+    "oracle",
+    "query",
+    "reader",
+    "rpc",
     "server",
     "store",
-    "structs",
+    "traversable",
+    "types",
 ]
-bundler = ["brk_bundler"]
+bencher = ["brk_bencher"]
+bindgen = ["brk_bindgen"]
+client = ["brk_client"]
 computer = ["brk_computer"]
 error = ["brk_error"]
 fetcher = ["brk_fetcher"]
+cohort = ["brk_cohort"]
 indexer = ["brk_indexer"]
-interface = ["brk_interface"]
+iterator = ["brk_iterator"]
 logger = ["brk_logger"]
-mcp = ["brk_mcp"]
-parser = ["brk_parser"]
+mempool = ["brk_mempool"]
+oracle = ["brk_oracle"]
+query = ["brk_query"]
+reader = ["brk_reader"]
+rpc = ["brk_rpc"]
 server = ["brk_server"]
 store = ["brk_store"]
-structs = ["brk_structs"]
+traversable = ["brk_traversable"]
+types = ["brk_types"]
 
 [dependencies]
-brk_bundler = { workspace = true, optional = true }
-brk_cli = { workspace = true }
+brk_bencher = { workspace = true, optional = true }
+brk_bindgen = { workspace = true, optional = true }
+brk_client = { workspace = true, optional = true }
 brk_computer = { workspace = true, optional = true }
 brk_error = { workspace = true, optional = true }
 brk_fetcher = { workspace = true, optional = true }
+brk_cohort = { workspace = true, optional = true }
 brk_indexer = { workspace = true, optional = true }
-brk_interface = { workspace = true, optional = true }
+brk_iterator = { workspace = true, optional = true }
 brk_logger = { workspace = true, optional = true }
-brk_mcp = { workspace = true, optional = true }
-brk_parser = { workspace = true, optional = true }
+brk_mempool = { workspace = true, optional = true }
+brk_oracle = { workspace = true, optional = true }
+brk_query = { workspace = true, optional = true }
+brk_reader = { workspace = true, optional = true }
+brk_rpc = { workspace = true, optional = true }
 brk_server = { workspace = true, optional = true }
 brk_store = { workspace = true, optional = true }
-brk_structs = { workspace = true, optional = true }
+brk_traversable = { workspace = true, optional = true }
+brk_types = { workspace = true, optional = true }
 
 [package.metadata.docs.rs]
 all-features = true

crates/brk/README.md

@@ -1,273 +1,68 @@
 # brk
 
-Unified Bitcoin Research Kit crate providing optional feature-gated access to all BRK components.
+Umbrella crate for the Bitcoin Research Kit.
 
-[![Crates.io](https://img.shields.io/crates/v/brk.svg)](https://crates.io/crates/brk)
-[![Documentation](https://docs.rs/brk/badge.svg)](https://docs.rs/brk)
+[crates.io](https://crates.io/crates/brk) | [docs.rs](https://docs.rs/brk)
 
-## Overview
+## Usage
 
-This crate serves as a unified entry point to the Bitcoin Research Kit ecosystem, providing feature-gated re-exports of all BRK components. It allows users to selectively include only the functionality they need while maintaining a single dependency declaration, with the `brk_cli` component always available for command-line interface access.
-
-**Key Features:**
-
-- Feature-gated modular access to 12 specialized BRK components
-- Single dependency entry point with selective compilation
-- Always-available CLI component for command-line operations
-- Comprehensive documentation aggregation with inline re-exports
-- `full` feature for complete BRK functionality inclusion
-- Optimized build configuration with docs.rs integration
-
-**Target Use Cases:**
-
-- Applications requiring selective BRK functionality to minimize dependencies
-- Library development where only specific Bitcoin analysis components are needed
-- Prototyping and experimentation with different BRK component combinations
-- Educational use cases demonstrating modular blockchain analytics architecture
-
-## Installation
+Single dependency to access any BRK component. Enable only what you need via feature flags.
 
 ```toml
 [dependencies]
-# Minimal installation with CLI only
-brk = "0.0.107"
-
-# Full functionality
-brk = { version = "0.0.107", features = ["full"] }
-
-# Selective features
-brk = { version = "0.0.107", features = ["indexer", "computer", "server"] }
+brk = { version = "0.1", features = ["query", "types"] }
 ```
 
-## Quick Start
-
-```rust
-// CLI is always available
-use brk::cli;
-
-// Feature-gated components
-#[cfg(feature = "indexer")]
-use brk::indexer::Indexer;
-
-#[cfg(feature = "computer")]
-use brk::computer::Computer;
-
-#[cfg(feature = "server")]
-use brk::server::Server;
-
-// Build complete pipeline with selected features
-#[cfg(all(feature = "indexer", feature = "computer", feature = "server"))]
-fn build_pipeline() -> Result<(), Box<dyn std::error::Error>> {
-    let indexer = Indexer::build("./data")?;
-    let computer = Computer::build("./analytics", &indexer)?;
-    let interface = brk::interface::Interface::build(&indexer, &computer);
-    let server = Server::new(interface, None);
-    Ok(())
-}
+```rust,ignore
+use brk::query::Query;
+use brk::types::Height;
 ```
 
-## API Overview
-
-### Feature Organization
-
-The crate provides feature-gated access to BRK components organized by functionality:
-
-**Core Data Processing:**
-- `structs` - Bitcoin-aware data structures and type system
-- `error` - Centralized error handling across components
-- `store` - Transactional key-value storage wrapper
-
-**Blockchain Processing:**
-- `parser` - Multi-threaded Bitcoin block parsing
-- `indexer` - Blockchain data indexing with columnar storage
-- `computer` - Analytics computation engine
-
-**Data Access:**
-- `interface` - Unified query interface with fuzzy search
-- `fetcher` - Multi-source price data aggregation
-
-**Service Layer:**
-- `server` - HTTP API server with caching and compression
-- `mcp` - Model Context Protocol bridge for LLM integration
-- `logger` - Enhanced logging with colored output
-
-**Web Infrastructure:**
-- `bundler` - Web asset bundling using Rolldown
-
-### Always Available
-
-**`cli`**: Command-line interface module (no feature gate required)
-Provides access to the complete BRK command-line interface for running full instances.
-
-## Examples
-
-### Minimal Bitcoin Parser
-
-```rust
-use brk::parser::Parser;
-
-fn parse_blocks() -> Result<(), Box<dyn std::error::Error>> {
-    #[cfg(feature = "parser")]
-    {
-        let parser = Parser::new("/path/to/blocks", None, rpc_client);
-        // Parse blockchain data
-        Ok(())
-    }
-    #[cfg(not(feature = "parser"))]
-    {
-        Err("Parser feature not enabled".into())
-    }
-}
-```
-
-### Analytics Pipeline
-
-```rust
-use brk::{indexer, computer, interface};
-
-#[cfg(all(feature = "indexer", feature = "computer", feature = "interface"))]
-fn analytics_pipeline() -> Result<(), Box<dyn std::error::Error>> {
-    // Initialize indexer
-    let indexer = indexer::Indexer::build("./blockchain_data")?;
-
-    // Compute analytics
-    let computer = computer::Computer::build("./analytics", &indexer)?;
-
-    // Create query interface
-    let interface = interface::Interface::build(&indexer, &computer);
-
-    // Query latest price
-    let params = interface::Params {
-        index: "date".to_string(),
-        ids: vec!["price-close".to_string()].into(),
-        from: Some(-1),
-        ..Default::default()
-    };
-
-    let result = interface.search_and_format(params)?;
-    println!("Latest Bitcoin price: {:?}", result);
-
-    Ok(())
-}
-```
-
-### Web Server Setup
-
-```rust
-use brk::{server, interface, indexer, computer};
-
-#[cfg(all(feature = "server", feature = "interface", feature = "indexer", feature = "computer"))]
-async fn web_server() -> Result<(), Box<dyn std::error::Error>> {
-    let indexer = indexer::Indexer::build("./data")?;
-    let computer = computer::Computer::build("./analytics", &indexer)?;
-    let interface = interface::Interface::build(&indexer, &computer);
-
-    let server = server::Server::new(interface, None);
-    server.serve(true).await?;
-
-    Ok(())
-}
-```
-
-### MCP Integration
-
-```rust
-use brk::{mcp, interface, indexer, computer};
-
-#[cfg(all(feature = "mcp", feature = "interface"))]
-fn mcp_server(interface: &'static interface::Interface) -> mcp::MCP {
-    mcp::MCP::new(interface)
-}
-```
-
-## Feature Combinations
-
-### Common Combinations
-
-**Data Processing**: `["structs", "parser", "indexer"]`
-Basic blockchain data processing and indexing.
-
-**Analytics**: `["indexer", "computer", "fetcher", "interface"]`
-Complete analytics pipeline with price data integration.
-
-**API Server**: `["interface", "server", "logger"]`
-HTTP API server with logging capabilities.
-
-**Full Stack**: `["full"]`
-All components for complete BRK functionality.
-
-### Dependency Optimization
-
-Feature selection allows for significant dependency reduction:
+Feature flags match crate names without the `brk_` prefix. Use `full` to enable all:
 
 ```toml
-# Minimal parser-only dependency
-brk = { version = "0.0.107", features = ["parser"] }
-
-# Analytics without web server
-brk = { version = "0.0.107", features = ["indexer", "computer", "interface"] }
-
-# Web server without parsing
-brk = { version = "0.0.107", features = ["interface", "server"] }
+[dependencies]
+brk = { version = "0.1", features = ["full"] }
 ```
 
-## Architecture
+## Crates
 
-### Re-export Pattern
+**Core Pipeline**
 
-The crate uses `#[doc(inline)]` re-exports to provide seamless access to component APIs:
+| Crate | Description |
+|-------|-------------|
+| [brk_reader](https://docs.rs/brk_reader) | Read blocks from `blk*.dat` with parallel parsing and XOR decoding |
+| [brk_indexer](https://docs.rs/brk_indexer) | Index transactions, addresses, and UTXOs |
+| [brk_computer](https://docs.rs/brk_computer) | Compute derived metrics (realized cap, MVRV, SOPR, cohorts, etc.) |
+| [brk_mempool](https://docs.rs/brk_mempool) | Monitor mempool, estimate fees, project upcoming blocks |
+| [brk_oracle](https://docs.rs/brk_oracle) | Pure on-chain BTC/USD price oracle |
+| [brk_query](https://docs.rs/brk_query) | Query interface for indexed and computed data |
+| [brk_server](https://docs.rs/brk_server) | REST API with OpenAPI docs |
 
-```rust
-#[cfg(feature = "component")]
-#[doc(inline)]
-pub use brk_component as component;
-```
+**Data & Storage**
 
-This pattern ensures:
-- Feature-gated compilation for dependency optimization
-- Inline documentation for unified API reference
-- Namespace preservation for component-specific functionality
+| Crate | Description |
+|-------|-------------|
+| [brk_types](https://docs.rs/brk_types) | Domain types: `Height`, `Sats`, `Txid`, addresses, etc. |
+| [brk_store](https://docs.rs/brk_store) | Key-value storage (fjall wrapper) |
+| [brk_fetcher](https://docs.rs/brk_fetcher) | Fetch price data from exchanges |
+| [brk_rpc](https://docs.rs/brk_rpc) | Bitcoin Core RPC client |
+| [brk_iterator](https://docs.rs/brk_iterator) | Unified block iteration with automatic source selection |
+| [brk_cohort](https://docs.rs/brk_cohort) | UTXO and address cohort filtering |
+| [brk_traversable](https://docs.rs/brk_traversable) | Navigate hierarchical data structures |
 
-### Build Configuration
+**Clients & Integration**
 
-- **Documentation**: `all-features = true` for complete docs.rs documentation
-- **CLI Integration**: `brk_cli` always available without feature gates
-- **Optional Dependencies**: All components except CLI are optional
+| Crate | Description |
+|-------|-------------|
+| [brk_client](https://docs.rs/brk_client) | Generated Rust API client |
+| [brk_bindgen](https://docs.rs/brk_bindgen) | Generate typed clients (Rust, JavaScript, Python) |
 
-## Configuration
+**Internal**
 
-### Feature Flags
-
-| Feature | Component | Description |
-|---------|-----------|-------------|
-| `bundler` | `brk_bundler` | Web asset bundling |
-| `computer` | `brk_computer` | Analytics computation |
-| `error` | `brk_error` | Error handling |
-| `fetcher` | `brk_fetcher` | Price data fetching |
-| `indexer` | `brk_indexer` | Blockchain indexing |
-| `interface` | `brk_interface` | Data query interface |
-| `logger` | `brk_logger` | Enhanced logging |
-| `mcp` | `brk_mcp` | Model Context Protocol |
-| `parser` | `brk_parser` | Block parsing |
-| `server` | `brk_server` | HTTP server |
-| `store` | `brk_store` | Key-value storage |
-| `structs` | `brk_structs` | Data structures |
-| `full` | All components | Complete functionality |
-
-### Documentation
-
-Documentation is aggregated from all components with `#![doc = include_str!("../README.md")]` ensuring comprehensive API reference across all features.
-
-## Code Analysis Summary
-
-**Main Structure**: Feature-gated re-export crate providing unified access to 12 BRK components \
-**Feature System**: Cargo features enabling selective compilation and dependency optimization \
-**CLI Integration**: Always-available `brk_cli` access without feature requirements \
-**Documentation**: Inline re-exports with comprehensive docs.rs integration \
-**Dependency Management**: Optional dependencies for all components except CLI \
-**Build Configuration**: Optimized compilation with all-features documentation \
-**Architecture**: Modular aggregation crate enabling flexible BRK ecosystem usage
-
----
-
-_This README was generated by Claude Code_
+| Crate | Description |
+|-------|-------------|
+| [brk_cli](https://docs.rs/brk_cli) | CLI binary (`cargo install --locked brk_cli`) |
+| [brk_error](https://docs.rs/brk_error) | Error types |
+| [brk_logger](https://docs.rs/brk_logger) | Logging infrastructure |
+| [brk_bencher](https://docs.rs/brk_bencher) | Benchmarking utilities |

crates/brk/build.rs

@@ -1,8 +0,0 @@
-fn main() {
-    let profile = std::env::var("PROFILE").unwrap_or_default();
-
-    if profile == "release" {
-        println!("cargo:rustc-flag=-C");
-        println!("cargo:rustc-flag=target-cpu=native");
-    }
-}

crates/brk/src/lib.rs

@@ -1,15 +1,20 @@
 #![doc = include_str!("../README.md")]
 
-#[cfg(feature = "bundler")]
+#[cfg(feature = "bencher")]
 #[doc(inline)]
-pub use brk_bundler as bundler;
+pub use brk_bencher as bencher;
 
+#[cfg(feature = "bindgen")]
 #[doc(inline)]
-pub use brk_cli as cli;
+pub use brk_bindgen as bindgen;
 
-#[cfg(feature = "structs")]
+#[cfg(feature = "client")]
 #[doc(inline)]
-pub use brk_structs as structs;
+pub use brk_client as client;
+
+#[cfg(feature = "cohort")]
+#[doc(inline)]
+pub use brk_cohort as cohort;
 
 #[cfg(feature = "computer")]
 #[doc(inline)]
@@ -27,21 +32,33 @@ pub use brk_fetcher as fetcher;
 #[doc(inline)]
 pub use brk_indexer as indexer;
 
+#[cfg(feature = "iterator")]
+#[doc(inline)]
+pub use brk_iterator as iterator;
+
 #[cfg(feature = "logger")]
 #[doc(inline)]
 pub use brk_logger as logger;
 
-#[cfg(feature = "mcp")]
+#[cfg(feature = "mempool")]
 #[doc(inline)]
-pub use brk_mcp as mcp;
+pub use brk_mempool as mempool;
 
-#[cfg(feature = "parser")]
+#[cfg(feature = "oracle")]
 #[doc(inline)]
-pub use brk_parser as parser;
+pub use brk_oracle as oracle;
 
-#[cfg(feature = "interface")]
+#[cfg(feature = "query")]
 #[doc(inline)]
-pub use brk_interface as interface;
+pub use brk_query as query;
+
+#[cfg(feature = "reader")]
+#[doc(inline)]
+pub use brk_reader as reader;
+
+#[cfg(feature = "rpc")]
+#[doc(inline)]
+pub use brk_rpc as rpc;
 
 #[cfg(feature = "server")]
 #[doc(inline)]
@@ -50,3 +67,11 @@ pub use brk_server as server;
 #[cfg(feature = "store")]
 #[doc(inline)]
 pub use brk_store as store;
+
+#[cfg(feature = "traversable")]
+#[doc(inline)]
+pub use brk_traversable as traversable;
+
+#[cfg(feature = "types")]
+#[doc(inline)]
+pub use brk_types as types;

crates/brk/src/main.rs

@@ -1 +0,0 @@
-use brk_cli::main;

crates/brk_alloc/Cargo.toml

@@ -0,0 +1,12 @@
+[package]
+name = "brk_alloc"
+description = "Global allocator and memory utilities for brk"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+homepage.workspace = true
+repository.workspace = true
+
+[dependencies]
+libmimalloc-sys = { version = "0.1.49", features = ["extended"] }
+mimalloc = { version = "0.1.52" }

crates/brk_alloc/src/lib.rs

@@ -0,0 +1,21 @@
+//! Global allocator and memory utilities for brk.
+//!
+//! This crate sets mimalloc as the global allocator and provides
+//! utilities for monitoring and managing memory.
+
+use mimalloc::MiMalloc as Allocator;
+
+#[global_allocator]
+static GLOBAL: Allocator = Allocator;
+
+/// Mimalloc allocator utilities
+pub struct Mimalloc;
+
+impl Mimalloc {
+    /// Eagerly free memory back to OS.
+    /// Only call at natural pause points.
+    #[inline]
+    pub fn collect() {
+        unsafe { libmimalloc_sys::mi_collect(true) }
+    }
+}

crates/brk_bencher/Cargo.toml

@@ -0,0 +1,16 @@
+[package]
+name = "brk_bencher"
+description = "A simple benchmarker for testing other crates."
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+homepage.workspace = true
+repository.workspace = true
+
+[dependencies]
+brk_error = { workspace = true }
+brk_logger = { workspace = true }
+parking_lot = { workspace = true }
+
+[target.'cfg(target_os = "macos")'.dependencies]
+libproc = "0.14"

crates/brk_bencher/README.md

@@ -0,0 +1,43 @@
+# brk_bencher
+
+Resource monitoring for long-running Bitcoin indexing operations.
+
+## What It Enables
+
+Track disk usage, memory consumption (current + peak), and I/O throughput during indexing runs. Progress tracking hooks into brk_logger to record processing milestones automatically.
+
+## Key Features
+
+- **Multi-metric monitoring**: Disk, memory (RSS + peak), I/O read/write
+- **Progress tracking**: Integrates with logging to capture block heights as they're processed
+- **Run comparison**: Outputs timestamped CSVs for comparing multiple runs
+- **macOS optimized**: Uses libproc for accurate process metrics on macOS
+- **Non-blocking**: Monitors in background thread with 5-second sample interval
+
+## Core API
+
+```rust,ignore
+let mut bencher = Bencher::from_cargo_env("brk_indexer", &data_path)?;
+bencher.start()?;
+
+// ... run indexing ...
+
+bencher.stop()?;
+```
+
+## Output Structure
+
+```
+benches/
+└── brk_indexer/
+    └── 1703001234/
+        ├── disk.csv      # timestamp_ms, bytes
+        ├── memory.csv    # timestamp_ms, current, peak
+        ├── io.csv        # timestamp_ms, read, written
+        └── progress.csv  # timestamp_ms, height
+```
+
+## Built On
+
+- `brk_error` for error handling
+- `brk_logger` for progress hook integration

crates/brk_bencher/src/disk.rs

@@ -0,0 +1,66 @@
+use std::{
+    collections::HashMap,
+    fs::{self, File},
+    io::{self, Write},
+    os::unix::fs::MetadataExt,
+    path::{Path, PathBuf},
+    time::SystemTime,
+};
+
+pub struct DiskMonitor {
+    cache: HashMap<PathBuf, (u64, SystemTime)>, // path -> (bytes_used, mtime)
+    monitored_path: PathBuf,
+    writer: File,
+}
+
+impl DiskMonitor {
+    pub fn new(monitored_path: &Path, csv_path: &Path) -> io::Result<Self> {
+        let mut writer = File::create(csv_path)?;
+        writeln!(writer, "timestamp_ms,disk_usage")?;
+
+        Ok(Self {
+            cache: HashMap::new(),
+            monitored_path: monitored_path.to_path_buf(),
+            writer,
+        })
+    }
+
+    /// Record disk usage at the given timestamp
+    pub fn record(&mut self, elapsed_ms: u128) -> io::Result<()> {
+        if let Ok(bytes) = self.scan_recursive(&self.monitored_path.clone()) {
+            writeln!(self.writer, "{},{}", elapsed_ms, bytes)?;
+        }
+        Ok(())
+    }
+
+    fn scan_recursive(&mut self, path: &Path) -> io::Result<u64> {
+        let mut total = 0;
+
+        for entry in fs::read_dir(path)? {
+            let entry = entry?;
+            let path = entry.path();
+            let metadata = entry.metadata()?;
+
+            if metadata.is_file() {
+                let mtime = metadata.modified()?;
+
+                // Check cache: if mtime unchanged, use cached value
+                if let Some((cached_bytes, cached_mtime)) = self.cache.get(&path)
+                    && *cached_mtime == mtime
+                {
+                    total += cached_bytes;
+                    continue;
+                }
+
+                // File is new or modified - get actual disk usage
+                let bytes = metadata.blocks() * 512;
+                self.cache.insert(path, (bytes, mtime));
+                total += bytes;
+            } else if metadata.is_dir() {
+                total += self.scan_recursive(&path)?;
+            }
+        }
+
+        Ok(total)
+    }
+}

crates/brk_bencher/src/io.rs

@@ -0,0 +1,83 @@
+use std::{
+    fs::File,
+    io::{self, Write},
+    path::Path,
+};
+
+#[cfg(target_os = "linux")]
+use std::fs;
+
+#[cfg(target_os = "macos")]
+use libproc::pid_rusage::{RUsageInfoV2, pidrusage};
+
+pub struct IoMonitor {
+    pid: u32,
+    writer: File,
+}
+
+impl IoMonitor {
+    pub fn new(pid: u32, csv_path: &Path) -> io::Result<Self> {
+        let mut writer = File::create(csv_path)?;
+        writeln!(writer, "timestamp_ms,bytes_read,bytes_written")?;
+
+        Ok(Self { pid, writer })
+    }
+
+    /// Record I/O usage at the given timestamp
+    pub fn record(&mut self, elapsed_ms: u128) -> io::Result<()> {
+        if let Ok((read, written)) = self.get_io_usage() {
+            writeln!(self.writer, "{},{},{}", elapsed_ms, read, written)?;
+        }
+        Ok(())
+    }
+
+    /// Get I/O usage in bytes
+    /// Returns (bytes_read, bytes_written)
+    fn get_io_usage(&self) -> io::Result<(u64, u64)> {
+        #[cfg(target_os = "linux")]
+        {
+            self.get_io_usage_linux()
+        }
+
+        #[cfg(target_os = "macos")]
+        {
+            self.get_io_usage_macos()
+        }
+    }
+
+    #[cfg(target_os = "linux")]
+    fn get_io_usage_linux(&self) -> io::Result<(u64, u64)> {
+        let io_content = fs::read_to_string(format!("/proc/{}/io", self.pid))?;
+
+        let mut read_bytes = None;
+        let mut write_bytes = None;
+
+        for line in io_content.lines() {
+            if line.starts_with("read_bytes:") {
+                if let Some(value_str) = line.split_whitespace().nth(1) {
+                    read_bytes = value_str.parse::<u64>().ok();
+                }
+            } else if line.starts_with("write_bytes:") {
+                if let Some(value_str) = line.split_whitespace().nth(1) {
+                    write_bytes = value_str.parse::<u64>().ok();
+                }
+            }
+        }
+
+        match (read_bytes, write_bytes) {
+            (Some(r), Some(w)) => Ok((r, w)),
+            _ => Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                "Failed to parse I/O stats from /proc/[pid]/io",
+            )),
+        }
+    }
+
+    #[cfg(target_os = "macos")]
+    fn get_io_usage_macos(&self) -> io::Result<(u64, u64)> {
+        match pidrusage::<RUsageInfoV2>(self.pid as i32) {
+            Ok(info) => Ok((info.ri_diskio_bytesread, info.ri_diskio_byteswritten)),
+            Err(_) => Err(io::Error::other("Failed to get process I/O stats")),
+        }
+    }
+}

crates/brk_bencher/src/lib.rs

@@ -0,0 +1,162 @@
+use std::{
+    fs,
+    path::{Path, PathBuf},
+    sync::{
+        Arc,
+        atomic::{AtomicBool, Ordering},
+    },
+    thread::{self, JoinHandle},
+    time::{Duration, Instant, SystemTime, UNIX_EPOCH},
+};
+
+use brk_error::{Error, Result};
+
+mod disk;
+mod io;
+mod memory;
+mod progression;
+
+use disk::*;
+use io::*;
+use memory::*;
+use parking_lot::Mutex;
+use progression::*;
+
+#[derive(Clone)]
+pub struct Bencher(Arc<BencherInner>);
+
+struct BencherInner {
+    bench_dir: PathBuf,
+    monitored_path: PathBuf,
+    stop_flag: Arc<AtomicBool>,
+    monitor_thread: Mutex<Option<JoinHandle<Result<()>>>>,
+    progression: Arc<ProgressionMonitor>,
+}
+
+impl Bencher {
+    /// Create a new bencher for the given crate name
+    /// Creates directory structure: workspace_root/benches/{crate_name}/{timestamp}/
+    pub fn new(crate_name: &str, workspace_root: &Path, monitored_path: &Path) -> Result<Self> {
+        let timestamp = SystemTime::now().duration_since(UNIX_EPOCH)?.as_secs();
+
+        let bench_dir = workspace_root
+            .join("benches")
+            .join(crate_name)
+            .join(timestamp.to_string());
+
+        fs::create_dir_all(&bench_dir)?;
+
+        let progress_csv = bench_dir.join("progress.csv");
+        let progression = Arc::new(ProgressionMonitor::new(&progress_csv)?);
+        let progression_clone = progression.clone();
+
+        // Register hook with logger
+        brk_logger::register_hook(move |message| {
+            progression_clone.check_and_record(message);
+        })
+        .map_err(|e| std::io::Error::new(std::io::ErrorKind::AlreadyExists, e))?;
+
+        Ok(Self(Arc::new(BencherInner {
+            bench_dir,
+            monitored_path: monitored_path.to_path_buf(),
+            stop_flag: Arc::new(AtomicBool::new(false)),
+            progression,
+            monitor_thread: Mutex::new(None),
+        })))
+    }
+
+    /// Create a bencher using CARGO_MANIFEST_DIR to find workspace root
+    pub fn from_cargo_env(crate_name: &str, monitored_path: &Path) -> Result<Self> {
+        let mut current = std::env::current_dir()
+            .map_err(|e| format!("Failed to get current directory: {}", e))
+            .unwrap();
+
+        let workspace_root = loop {
+            let cargo_toml = current.join("Cargo.toml");
+            if cargo_toml.exists() {
+                let contents = std::fs::read_to_string(&cargo_toml)
+                    .map_err(|e| format!("Failed to read Cargo.toml: {}", e))
+                    .unwrap();
+                if contents.contains("[workspace]") {
+                    break current;
+                }
+            }
+
+            current = current
+                .parent()
+                .ok_or(Error::NotFound("Workspace root not found".into()))?
+                .to_path_buf();
+        };
+
+        Self::new(crate_name, &workspace_root, monitored_path)
+    }
+
+    /// Start monitoring disk usage and memory footprint
+    pub fn start(&mut self) -> Result<()> {
+        if self.0.monitor_thread.lock().is_some() {
+            return Err(Error::Internal("Bencher already started"));
+        }
+
+        let stop_flag = self.0.stop_flag.clone();
+        let bench_dir = self.0.bench_dir.clone();
+        let monitored_path = self.0.monitored_path.clone();
+
+        let handle =
+            thread::spawn(move || monitor_resources(&monitored_path, &bench_dir, stop_flag));
+
+        *self.0.monitor_thread.lock() = Some(handle);
+        Ok(())
+    }
+
+    /// Stop monitoring and wait for the thread to finish
+    pub fn stop(&self) -> Result<()> {
+        self.0.stop_flag.store(true, Ordering::Relaxed);
+
+        if let Some(handle) = self.0.monitor_thread.lock().take() {
+            handle
+                .join()
+                .map_err(|_| Error::Internal("Monitor thread panicked"))??;
+        }
+
+        self.0.progression.flush()?;
+
+        Ok(())
+    }
+}
+
+impl Drop for Bencher {
+    fn drop(&mut self) {
+        let _ = self.stop();
+    }
+}
+
+fn monitor_resources(
+    monitored_path: &Path,
+    bench_dir: &Path,
+    stop_flag: Arc<AtomicBool>,
+) -> Result<()> {
+    let pid = std::process::id();
+    let start = Instant::now();
+
+    let mut disk_monitor = DiskMonitor::new(monitored_path, &bench_dir.join("disk.csv"))?;
+    let mut memory_monitor = MemoryMonitor::new(pid, &bench_dir.join("memory.csv"))?;
+    let mut io_monitor = IoMonitor::new(pid, &bench_dir.join("io.csv"))?;
+
+    'l: loop {
+        let elapsed_ms = start.elapsed().as_millis();
+
+        disk_monitor.record(elapsed_ms)?;
+        memory_monitor.record(elapsed_ms)?;
+        io_monitor.record(elapsed_ms)?;
+
+        for _ in 0..50 {
+            // 50 * 100ms = 5 seconds
+            if stop_flag.load(Ordering::Relaxed) {
+                break 'l;
+            }
+            thread::sleep(Duration::from_millis(100));
+        }
+    }
+
+    Ok(())
+}

crates/brk_bencher/src/memory.rs

@@ -0,0 +1,143 @@
+use std::{
+    fs::File,
+    io::{self, Write},
+    path::Path,
+};
+
+#[cfg(target_os = "linux")]
+use std::fs;
+
+#[cfg(target_os = "macos")]
+use std::process::Command;
+
+pub struct MemoryMonitor {
+    pid: u32,
+    writer: File,
+}
+
+impl MemoryMonitor {
+    pub fn new(pid: u32, csv_path: &Path) -> io::Result<Self> {
+        let mut writer = File::create(csv_path)?;
+        writeln!(writer, "timestamp_ms,phys_footprint,phys_footprint_peak")?;
+
+        Ok(Self { pid, writer })
+    }
+
+    /// Record memory usage at the given timestamp
+    pub fn record(&mut self, elapsed_ms: u128) -> io::Result<()> {
+        if let Ok((footprint, peak)) = self.get_memory_usage() {
+            writeln!(self.writer, "{},{},{}", elapsed_ms, footprint, peak)?;
+        }
+        Ok(())
+    }
+
+    /// Get memory usage in bytes
+    /// Returns (current_bytes, peak_bytes)
+    fn get_memory_usage(&self) -> io::Result<(u64, u64)> {
+        #[cfg(target_os = "linux")]
+        {
+            self.get_memory_usage_linux()
+        }
+
+        #[cfg(target_os = "macos")]
+        {
+            self.get_memory_usage_macos()
+        }
+    }
+
+    #[cfg(target_os = "linux")]
+    fn get_memory_usage_linux(&self) -> io::Result<(u64, u64)> {
+        let status_content = fs::read_to_string(format!("/proc/{}/status", self.pid))?;
+
+        let mut vm_rss = None;
+        let mut vm_hwm = None;
+
+        for line in status_content.lines() {
+            if line.starts_with("VmRSS:") {
+                if let Some(value_str) = line.split_whitespace().nth(1) {
+                    if let Ok(kb) = value_str.parse::<u64>() {
+                        vm_rss = Some(kb * 1024); // KiB to bytes
+                    }
+                }
+            } else if line.starts_with("VmHWM:") {
+                if let Some(value_str) = line.split_whitespace().nth(1) {
+                    if let Ok(kb) = value_str.parse::<u64>() {
+                        vm_hwm = Some(kb * 1024); // KiB to bytes
+                    }
+                }
+            }
+        }
+
+        match (vm_rss, vm_hwm) {
+            (Some(rss), Some(hwm)) => Ok((rss, hwm)),
+            _ => Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                "Failed to parse memory info from /proc/[pid]/status",
+            )),
+        }
+    }
+
+    #[cfg(target_os = "macos")]
+    fn get_memory_usage_macos(&self) -> io::Result<(u64, u64)> {
+        let output = Command::new("footprint")
+            .args(["-p", &self.pid.to_string()])
+            .output()?;
+
+        let stdout = String::from_utf8(output.stdout).map_err(|_| {
+            io::Error::new(io::ErrorKind::InvalidData, "Invalid UTF-8 from footprint")
+        })?;
+
+        parse_footprint_output(&stdout).ok_or_else(|| {
+            io::Error::new(
+                io::ErrorKind::InvalidData,
+                "Failed to parse footprint output",
+            )
+        })
+    }
+}
+
+#[cfg(target_os = "macos")]
+fn parse_footprint_output(output: &str) -> Option<(u64, u64)> {
+    let mut phys_footprint = None;
+    let mut phys_footprint_peak = None;
+
+    for line in output.lines() {
+        let line = line.trim();
+
+        if line.starts_with("phys_footprint:") {
+            // Format: "phys_footprint: 7072 KB"
+            let parts: Vec<&str> = line.split_whitespace().collect();
+            if parts.len() >= 3 {
+                // parts[0] = "phys_footprint:"
+                // parts[1] = "7072"
+                // parts[2] = "KB"
+                phys_footprint = parse_size_to_bytes(parts[1], parts[2]);
+            }
+        } else if line.starts_with("phys_footprint_peak:") {
+            // Format: "phys_footprint_peak: 15 MB"
+            let parts: Vec<&str> = line.split_whitespace().collect();
+            if parts.len() >= 3 {
+                phys_footprint_peak = parse_size_to_bytes(parts[1], parts[2]);
+            }
+        }
+    }
+
+    match (phys_footprint, phys_footprint_peak) {
+        (Some(f), Some(p)) => Some((f, p)),
+        _ => None,
+    }
+}
+
+#[cfg(target_os = "macos")]
+fn parse_size_to_bytes(value: &str, unit: &str) -> Option<u64> {
+    let value: f64 = value.parse().ok()?;
+
+    let multiplier = match unit.to_uppercase().as_str() {
+        "KB" => 1024.0,                   // KiB to bytes
+        "MB" => 1024.0 * 1024.0,          // MiB to bytes
+        "GB" => 1024.0 * 1024.0 * 1024.0, // GiB to bytes
+        _ => return None,
+    };
+
+    Some((value * multiplier) as u64)
+}

crates/brk_bencher/src/progression.rs

@@ -0,0 +1,74 @@
+use parking_lot::Mutex;
+use std::{
+    fs,
+    io::{self, BufWriter, Write},
+    path::Path,
+    time::Instant,
+};
+
+/// Patterns to match for progress tracking.
+const PROGRESS_PATTERNS: &[&str] = &[
+    "block ",    // "Indexing block 123..."
+    "chain at ", // "Processing chain at 456..."
+];
+
+pub struct ProgressionMonitor {
+    csv_file: Mutex<BufWriter<fs::File>>,
+    start_time: Instant,
+}
+
+impl ProgressionMonitor {
+    pub fn new(csv_path: &Path) -> io::Result<Self> {
+        let mut csv_file = BufWriter::new(fs::File::create(csv_path)?);
+        writeln!(csv_file, "timestamp_ms,value")?;
+
+        Ok(Self {
+            csv_file: Mutex::new(csv_file),
+            start_time: Instant::now(),
+        })
+    }
+
+    /// Check message for progress patterns and record if found
+    #[inline]
+    pub fn check_and_record(&self, message: &str) {
+        let Some(value) = parse_progress(message) else {
+            return;
+        };
+
+        if value % 10 != 0 {
+            return;
+        }
+
+        let elapsed_ms = self.start_time.elapsed().as_millis();
+        let _ = writeln!(self.csv_file.lock(), "{},{}", elapsed_ms, value);
+    }
+
+    pub fn flush(&self) -> io::Result<()> {
+        self.csv_file.lock().flush()
+    }
+}
+
+/// Parse progress value from message
+#[inline]
+fn parse_progress(message: &str) -> Option<u64> {
+    PROGRESS_PATTERNS
+        .iter()
+        .find_map(|pattern| parse_number_after(message, pattern))
+}
+
+/// Extract number immediately following the pattern
+#[inline]
+fn parse_number_after(message: &str, pattern: &str) -> Option<u64> {
+    let start = message.find(pattern)?;
+    let after = &message[start + pattern.len()..];
+
+    let end = after
+        .find(|c: char| !c.is_ascii_digit())
+        .unwrap_or(after.len());
+
+    if end == 0 {
+        return None;
+    }
+
+    after[..end].parse().ok()
+}

crates/brk_bencher_visualizer/Cargo.toml

@@ -0,0 +1,12 @@
+[package]
+name = "brk_bencher_visualizer"
+description = "A generator of charts for brk_bencher"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+homepage.workspace = true
+repository.workspace = true
+publish = false
+
+[dependencies]
+plotters = "0.3.7"

crates/brk_bencher_visualizer/README.md

@@ -0,0 +1,34 @@
+# brk_bencher_visualizer
+
+SVG chart generation for benchmark visualization.
+
+## What It Enables
+
+Turn benchmark CSV data into publication-ready SVG charts showing disk usage, memory (current/peak), progress, and I/O over time. Compare multiple runs side-by-side with automatic color coding.
+
+## Key Features
+
+- **Multi-run comparison**: Overlay multiple benchmark runs with distinct colors
+- **Dual-axis charts**: Memory charts show both current and peak usage (solid vs dashed lines)
+- **Smart scaling**: Automatic unit conversion for bytes (KB/MB/GB) and time (seconds/minutes/hours)
+- **Per-run trimming**: Aligns data by progress cutoffs for fair comparison
+- **Dark theme**: Clean, readable charts with monospace fonts
+
+## Core API
+
+```rust,ignore
+let viz = Visualizer::from_cargo_env()?;
+viz.generate_all_charts()?;  // Process all crates in benches/
+```
+
+## Chart Types
+
+- `disk.svg` - Storage consumption over time
+- `memory.svg` - Current + peak memory usage
+- `progress.svg` - Processing progress (e.g., blocks indexed)
+- `io_read.svg` / `io_write.svg` - I/O throughput
+
+## Input Format
+
+Reads CSV files from `benches/<crate>/<run_id>/`:
+- `disk.csv`, `memory.csv`, `progress.csv`, `io.csv`

crates/brk_bencher_visualizer/src/chart.rs

@@ -0,0 +1,285 @@
+use crate::data::{DataPoint, DualRun, Result, Run};
+use crate::format;
+use plotters::prelude::*;
+use std::path::Path;
+
+const FONT: &str = "monospace";
+const FONT_SIZE: i32 = 20;
+const FONT_SIZE_BIG: i32 = 30;
+const SIZE: (u32, u32) = (2000, 1000);
+const TIME_BUFFER_MS: u64 = 10_000;
+
+const BG_COLOR: RGBColor = RGBColor(18, 18, 24);
+const TEXT_COLOR: RGBColor = RGBColor(230, 230, 240);
+const COLORS: [RGBColor; 6] = [
+    RGBColor(255, 99, 132),  // Pink/Red
+    RGBColor(54, 162, 235),  // Blue
+    RGBColor(75, 192, 192),  // Teal
+    RGBColor(255, 206, 86),  // Yellow
+    RGBColor(153, 102, 255), // Purple
+    RGBColor(255, 159, 64),  // Orange
+];
+
+pub enum YAxisFormat {
+    Bytes,
+    Number,
+}
+
+pub struct ChartConfig<'a> {
+    pub output_path: &'a Path,
+    pub title: String,
+    pub y_label: String,
+    pub y_format: YAxisFormat,
+}
+
+/// Generate a simple line chart from runs
+pub fn generate(config: ChartConfig, runs: &[Run]) -> Result<()> {
+    if runs.is_empty() {
+        return Ok(());
+    }
+
+    let max_time_ms = runs.iter().map(|r| r.max_timestamp()).max().unwrap_or(1000) + TIME_BUFFER_MS;
+    let max_time_s = max_time_ms as f64 / 1000.0;
+    let max_value = runs.iter().map(|r| r.max_value()).fold(0.0, f64::max);
+
+    let (time_scaled, time_divisor, time_label) = format::time(max_time_s);
+    let (value_scaled, scale_factor, y_label) =
+        scale_y_axis(max_value, &config.y_label, &config.y_format);
+    let x_labels = label_count(time_scaled);
+
+    let root = SVGBackend::new(config.output_path, SIZE).into_drawing_area();
+    root.fill(&BG_COLOR)?;
+
+    let mut chart = ChartBuilder::on(&root)
+        .caption(
+            &config.title,
+            (FONT, FONT_SIZE_BIG).into_font().color(&TEXT_COLOR),
+        )
+        .margin(20)
+        .margin_right(40)
+        .x_label_area_size(50)
+        .margin_left(50)
+        .right_y_label_area_size(75)
+        .build_cartesian_2d(0.0..time_scaled * 1.025, 0.0..value_scaled * 1.1)?;
+
+    configure_mesh(&mut chart, time_label, &y_label, &config.y_format, x_labels)?;
+
+    for (idx, run) in runs.iter().enumerate() {
+        let color = COLORS[idx % COLORS.len()];
+        draw_series(
+            &mut chart,
+            &run.data,
+            &run.id,
+            color,
+            time_divisor,
+            scale_factor,
+        )?;
+    }
+
+    configure_legend(&mut chart)?;
+    root.present()?;
+    println!("Generated: {}", config.output_path.display());
+    Ok(())
+}
+
+/// Generate a chart with dual series per run (e.g., current + peak memory)
+pub fn generate_dual(
+    config: ChartConfig,
+    runs: &[DualRun],
+    primary_suffix: &str,
+    secondary_suffix: &str,
+) -> Result<()> {
+    if runs.is_empty() {
+        return Ok(());
+    }
+
+    let max_time_ms = runs
+        .iter()
+        .flat_map(|r| r.primary.iter().chain(r.secondary.iter()))
+        .map(|d| d.timestamp_ms)
+        .max()
+        .unwrap_or(1000)
+        + TIME_BUFFER_MS;
+    let max_time_s = max_time_ms as f64 / 1000.0;
+    let max_value = runs.iter().map(|r| r.max_value()).fold(0.0, f64::max);
+
+    let (time_scaled, time_divisor, time_label) = format::time(max_time_s);
+    let (value_scaled, scale_factor, y_label) =
+        scale_y_axis(max_value, &config.y_label, &config.y_format);
+    let x_labels = label_count(time_scaled);
+
+    let root = SVGBackend::new(config.output_path, SIZE).into_drawing_area();
+    root.fill(&BG_COLOR)?;
+
+    let mut chart = ChartBuilder::on(&root)
+        .caption(
+            &config.title,
+            (FONT, FONT_SIZE_BIG).into_font().color(&TEXT_COLOR),
+        )
+        .margin(20)
+        .margin_right(40)
+        .x_label_area_size(50)
+        .margin_left(50)
+        .right_y_label_area_size(75)
+        .build_cartesian_2d(0.0..time_scaled * 1.025, 0.0..value_scaled * 1.1)?;
+
+    configure_mesh(&mut chart, time_label, &y_label, &config.y_format, x_labels)?;
+
+    for (idx, run) in runs.iter().enumerate() {
+        let color = COLORS[idx % COLORS.len()];
+
+        // Primary series (solid)
+        draw_series(
+            &mut chart,
+            &run.primary,
+            &format!("{} {}", run.id, primary_suffix),
+            color,
+            time_divisor,
+            scale_factor,
+        )?;
+
+        // Secondary series (dashed)
+        draw_dashed_series(
+            &mut chart,
+            &run.secondary,
+            &format!("{} {}", run.id, secondary_suffix),
+            color.mix(0.5),
+            time_divisor,
+            scale_factor,
+        )?;
+    }
+
+    configure_legend(&mut chart)?;
+    root.present()?;
+    println!("Generated: {}", config.output_path.display());
+    Ok(())
+}
+
+fn scale_y_axis(max_value: f64, base_label: &str, y_format: &YAxisFormat) -> (f64, f64, String) {
+    match y_format {
+        YAxisFormat::Bytes => {
+            let (scaled, unit) = format::bytes(max_value);
+            let factor = max_value / scaled;
+            (scaled, factor, format!("{} ({})", base_label, unit))
+        }
+        YAxisFormat::Number => (max_value, 1.0, base_label.to_string()),
+    }
+}
+
+/// Calculate appropriate label count to avoid duplicates when rounding to integers
+fn label_count(max_value: f64) -> usize {
+    let max_int = max_value.ceil() as usize;
+    // Don't exceed the range, cap at 12 for readability
+    max_int.clamp(2, 12)
+}
+
+type Chart<'a, 'b> = ChartContext<
+    'a,
+    SVGBackend<'b>,
+    Cartesian2d<plotters::coord::types::RangedCoordf64, plotters::coord::types::RangedCoordf64>,
+>;
+
+fn configure_mesh(
+    chart: &mut Chart,
+    x_label: &str,
+    y_label: &str,
+    y_format: &YAxisFormat,
+    x_labels: usize,
+) -> Result<()> {
+    let y_formatter: Box<dyn Fn(&f64) -> String> = match y_format {
+        YAxisFormat::Bytes => Box::new(|y: &f64| {
+            if y.fract() == 0.0 {
+                format!("{:.0}", y)
+            } else {
+                format!("{:.1}", y)
+            }
+        }),
+        YAxisFormat::Number => Box::new(|y: &f64| format::axis_number(*y)),
+    };
+
+    chart
+        .configure_mesh()
+        .disable_mesh()
+        .x_desc(x_label)
+        .y_desc(y_label)
+        .x_label_formatter(&|x| format!("{:.0}", x))
+        .y_label_formatter(&y_formatter)
+        .x_labels(x_labels)
+        .y_labels(10)
+        .x_label_style((FONT, FONT_SIZE).into_font().color(&TEXT_COLOR.mix(0.7)))
+        .y_label_style((FONT, FONT_SIZE).into_font().color(&TEXT_COLOR.mix(0.7)))
+        .axis_style(TEXT_COLOR.mix(0.3))
+        .draw()?;
+    Ok(())
+}
+
+fn draw_series(
+    chart: &mut Chart,
+    data: &[DataPoint],
+    label: &str,
+    color: RGBColor,
+    time_divisor: f64,
+    scale_factor: f64,
+) -> Result<()> {
+    let points = data.iter().map(|d| {
+        (
+            d.timestamp_ms as f64 / 1000.0 / time_divisor,
+            d.value / scale_factor,
+        )
+    });
+
+    chart
+        .draw_series(LineSeries::new(points, color.stroke_width(1)))?
+        .label(label)
+        .legend(move |(x, y)| PathElement::new(vec![(x, y), (x + 20, y)], color.stroke_width(1)));
+    Ok(())
+}
+
+fn draw_dashed_series(
+    chart: &mut Chart,
+    data: &[DataPoint],
+    label: &str,
+    color: RGBAColor,
+    time_divisor: f64,
+    scale_factor: f64,
+) -> Result<()> {
+    let points: Vec<_> = data
+        .iter()
+        .map(|d| {
+            (
+                d.timestamp_ms as f64 / 1000.0 / time_divisor,
+                d.value / scale_factor,
+            )
+        })
+        .collect();
+
+    // Draw dashed line by skipping every other segment
+    chart
+        .draw_series(
+            points
+                .windows(2)
+                .enumerate()
+                .filter(|(i, _)| i % 2 == 0)
+                .map(|(_, w)| PathElement::new(vec![w[0], w[1]], color.stroke_width(2))),
+        )?
+        .label(label)
+        .legend(move |(x, y)| {
+            PathElement::new(
+                vec![(x, y), (x + 10, y), (x + 20, y)],
+                color.stroke_width(2),
+            )
+        });
+    Ok(())
+}
+
+fn configure_legend<'a>(chart: &mut Chart<'a, 'a>) -> Result<()> {
+    chart
+        .configure_series_labels()
+        .position(SeriesLabelPosition::UpperLeft)
+        .label_font((FONT, FONT_SIZE).into_font().color(&TEXT_COLOR.mix(0.9)))
+        .background_style(BG_COLOR.mix(0.98))
+        .border_style(BG_COLOR)
+        .margin(10)
+        .draw()?;
+    Ok(())
+}

crates/brk_bencher_visualizer/src/data.rs

@@ -0,0 +1,238 @@
+use std::{collections::HashMap, fs, path::Path};
+
+pub type Result<T> = std::result::Result<T, Box<dyn std::error::Error>>;
+
+#[derive(Debug, Clone)]
+pub struct DataPoint {
+    pub timestamp_ms: u64,
+    pub value: f64,
+}
+
+/// Per-run cutoff timestamps for fair comparison
+pub struct Cutoffs {
+    by_id: HashMap<String, u64>,
+    default: u64,
+}
+
+impl Cutoffs {
+    /// Calculate cutoffs from progress runs.
+    /// Finds the common max progress, then returns when each run reached it.
+    pub fn from_progress(progress_runs: &[Run]) -> Self {
+        const TIME_BUFFER_MS: u64 = 10_000;
+
+        if progress_runs.is_empty() {
+            return Self {
+                by_id: HashMap::new(),
+                default: u64::MAX,
+            };
+        }
+
+        // Find the minimum of max progress values (the common point all runs reached)
+        let common_progress = progress_runs
+            .iter()
+            .map(|r| r.max_value())
+            .fold(f64::MAX, f64::min);
+
+        let by_id: HashMap<_, _> = progress_runs
+            .iter()
+            .map(|run| {
+                let cutoff = run
+                    .data
+                    .iter()
+                    .find(|d| d.value >= common_progress)
+                    .map(|d| d.timestamp_ms)
+                    .unwrap_or_else(|| run.max_timestamp())
+                    .saturating_add(TIME_BUFFER_MS);
+                (run.id.clone(), cutoff)
+            })
+            .collect();
+
+        let default = by_id.values().copied().max().unwrap_or(u64::MAX);
+
+        Self { by_id, default }
+    }
+
+    pub fn get(&self, id: &str) -> u64 {
+        self.by_id.get(id).copied().unwrap_or(self.default)
+    }
+
+    pub fn trim_runs(&self, runs: &[Run]) -> Vec<Run> {
+        runs.iter().map(|r| r.trimmed(self.get(&r.id))).collect()
+    }
+
+    pub fn trim_dual_runs(&self, runs: &[DualRun]) -> Vec<DualRun> {
+        runs.iter().map(|r| r.trimmed(self.get(&r.id))).collect()
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct Run {
+    pub id: String,
+    pub data: Vec<DataPoint>,
+}
+
+impl Run {
+    pub fn max_timestamp(&self) -> u64 {
+        self.data.iter().map(|d| d.timestamp_ms).max().unwrap_or(0)
+    }
+
+    pub fn max_value(&self) -> f64 {
+        self.data.iter().map(|d| d.value).fold(0.0, f64::max)
+    }
+
+    pub fn trimmed(&self, max_timestamp_ms: u64) -> Self {
+        Self {
+            id: self.id.clone(),
+            data: self
+                .data
+                .iter()
+                .filter(|d| d.timestamp_ms <= max_timestamp_ms)
+                .cloned()
+                .collect(),
+        }
+    }
+}
+
+/// Two data series from a single run (e.g., memory footprint + peak, or io read + write)
+#[derive(Debug, Clone)]
+pub struct DualRun {
+    pub id: String,
+    pub primary: Vec<DataPoint>,
+    pub secondary: Vec<DataPoint>,
+}
+
+impl DualRun {
+    pub fn trimmed(&self, max_timestamp_ms: u64) -> Self {
+        Self {
+            id: self.id.clone(),
+            primary: self
+                .primary
+                .iter()
+                .filter(|d| d.timestamp_ms <= max_timestamp_ms)
+                .cloned()
+                .collect(),
+            secondary: self
+                .secondary
+                .iter()
+                .filter(|d| d.timestamp_ms <= max_timestamp_ms)
+                .cloned()
+                .collect(),
+        }
+    }
+
+    pub fn max_value(&self) -> f64 {
+        self.primary
+            .iter()
+            .chain(self.secondary.iter())
+            .map(|d| d.value)
+            .fold(0.0, f64::max)
+    }
+}
+
+pub fn read_runs(crate_path: &Path, filename: &str) -> Result<Vec<Run>> {
+    let mut runs = Vec::new();
+
+    for entry in fs::read_dir(crate_path)? {
+        let run_path = entry?.path();
+        if !run_path.is_dir() {
+            continue;
+        }
+
+        let run_id = run_path
+            .file_name()
+            .and_then(|n| n.to_str())
+            .ok_or("Invalid run ID")?
+            .to_string();
+
+        // Skip underscore-prefixed or numeric-only directories
+        if run_id.starts_with('_') || run_id.chars().all(|c| c.is_ascii_digit()) {
+            continue;
+        }
+
+        let csv_path = run_path.join(filename);
+        if csv_path.exists()
+            && let Ok(data) = read_csv(&csv_path)
+        {
+            runs.push(Run { id: run_id, data });
+        }
+    }
+
+    Ok(runs)
+}
+
+pub fn read_dual_runs(crate_path: &Path, filename: &str) -> Result<Vec<DualRun>> {
+    let mut runs = Vec::new();
+
+    for entry in fs::read_dir(crate_path)? {
+        let run_path = entry?.path();
+        if !run_path.is_dir() {
+            continue;
+        }
+
+        let run_id = run_path
+            .file_name()
+            .and_then(|n| n.to_str())
+            .ok_or("Invalid run ID")?
+            .to_string();
+
+        if run_id.starts_with('_') || run_id.chars().all(|c| c.is_ascii_digit()) {
+            continue;
+        }
+
+        let csv_path = run_path.join(filename);
+        if csv_path.exists()
+            && let Ok((primary, secondary)) = read_dual_csv(&csv_path)
+        {
+            runs.push(DualRun {
+                id: run_id,
+                primary,
+                secondary,
+            });
+        }
+    }
+
+    Ok(runs)
+}
+
+fn read_csv(path: &Path) -> Result<Vec<DataPoint>> {
+    let content = fs::read_to_string(path)?;
+    let data = content
+        .lines()
+        .skip(1) // header
+        .filter_map(|line| {
+            let mut parts = line.split(',');
+            let timestamp_ms = parts.next()?.parse().ok()?;
+            let value = parts.next()?.parse().ok()?;
+            Some(DataPoint {
+                timestamp_ms,
+                value,
+            })
+        })
+        .collect();
+    Ok(data)
+}
+
+fn read_dual_csv(path: &Path) -> Result<(Vec<DataPoint>, Vec<DataPoint>)> {
+    let content = fs::read_to_string(path)?;
+    let mut primary = Vec::new();
+    let mut secondary = Vec::new();
+
+    for line in content.lines().skip(1) {
+        let mut parts = line.split(',');
+        if let (Some(ts), Some(v1), Some(v2)) = (parts.next(), parts.next(), parts.next())
+            && let (Ok(timestamp_ms), Ok(val1), Ok(val2)) =
+                (ts.parse(), v1.parse::<f64>(), v2.parse::<f64>())
+        {
+            primary.push(DataPoint {
+                timestamp_ms,
+                value: val1,
+            });
+            secondary.push(DataPoint {
+                timestamp_ms,
+                value: val2,
+            });
+        }
+    }
+
+    Ok((primary, secondary))
+}

crates/brk_bencher_visualizer/src/format.rs

@@ -0,0 +1,45 @@
+const KIB: f64 = 1024.0;
+const MIB: f64 = KIB * 1024.0;
+const GIB: f64 = MIB * 1024.0;
+
+const MINUTE: f64 = 60.0;
+const HOUR: f64 = 3600.0;
+
+/// Returns (scaled_value, unit_suffix)
+pub fn bytes(bytes: f64) -> (f64, &'static str) {
+    if bytes >= GIB {
+        (bytes / GIB, "GiB")
+    } else if bytes >= MIB {
+        (bytes / MIB, "MiB")
+    } else if bytes >= KIB {
+        (bytes / KIB, "KiB")
+    } else {
+        (bytes, "bytes")
+    }
+}
+
+/// Returns (scaled_value, divisor, axis_label)
+pub fn time(seconds: f64) -> (f64, f64, &'static str) {
+    if seconds >= HOUR * 2.0 {
+        (seconds / HOUR, HOUR, "Time (h)")
+    } else if seconds >= MINUTE * 2.0 {
+        (seconds / MINUTE, MINUTE, "Time (min)")
+    } else {
+        (seconds, 1.0, "Time (s)")
+    }
+}
+
+pub fn axis_number(value: f64) -> String {
+    if value >= 1000.0 {
+        let k = value / 1000.0;
+        if k.fract() == 0.0 || k >= 100.0 {
+            format!("{:.0}k", k)
+        } else if k >= 10.0 {
+            format!("{:.1}k", k)
+        } else {
+            format!("{:.2}k", k)
+        }
+    } else {
+        format!("{:.0}", value)
+    }
+}

crates/brk_bencher_visualizer/src/lib.rs

@@ -0,0 +1,262 @@
+mod chart;
+mod data;
+mod format;
+
+use data::{Cutoffs, DualRun, Result, Run, read_dual_runs, read_runs};
+use std::{
+    fs,
+    path::{Path, PathBuf},
+};
+
+pub struct Visualizer {
+    workspace_root: PathBuf,
+}
+
+impl Visualizer {
+    pub fn new(workspace_root: impl AsRef<Path>) -> Self {
+        Self {
+            workspace_root: workspace_root.as_ref().to_path_buf(),
+        }
+    }
+
+    pub fn from_cargo_env() -> Result<Self> {
+        let workspace_root = Path::new(env!("CARGO_MANIFEST_DIR"))
+            .parent()
+            .and_then(|p| p.parent())
+            .ok_or("Failed to find workspace root")?
+            .to_path_buf();
+        Ok(Self { workspace_root })
+    }
+
+    pub fn generate_all_charts(&self) -> Result<()> {
+        let benches_dir = self.workspace_root.join("benches");
+        if !benches_dir.exists() {
+            return Err("Benches directory does not exist".into());
+        }
+
+        for entry in fs::read_dir(&benches_dir)? {
+            let path = entry?.path();
+            if path.is_dir() {
+                let crate_name = path
+                    .file_name()
+                    .and_then(|n| n.to_str())
+                    .ok_or("Invalid crate name")?;
+
+                println!("Generating charts for crate: {}", crate_name);
+                self.generate_crate_charts(&path, crate_name)?;
+            }
+        }
+
+        Ok(())
+    }
+
+    fn generate_crate_charts(&self, crate_path: &Path, crate_name: &str) -> Result<()> {
+        let disk_runs = read_runs(crate_path, "disk.csv")?;
+        let memory_runs = read_dual_runs(crate_path, "memory.csv")?;
+        let progress_runs = read_runs(crate_path, "progress.csv")?;
+        let io_runs = read_dual_runs(crate_path, "io.csv")?;
+
+        // Combined charts (all runs)
+        self.generate_combined_charts(
+            crate_path,
+            crate_name,
+            &disk_runs,
+            &memory_runs,
+            &progress_runs,
+            &io_runs,
+        )?;
+
+        // Individual charts (one per run)
+        self.generate_individual_charts(
+            crate_path,
+            crate_name,
+            &disk_runs,
+            &memory_runs,
+            &progress_runs,
+            &io_runs,
+        )?;
+
+        Ok(())
+    }
+
+    fn generate_combined_charts(
+        &self,
+        crate_path: &Path,
+        crate_name: &str,
+        disk_runs: &[Run],
+        memory_runs: &[DualRun],
+        progress_runs: &[Run],
+        io_runs: &[DualRun],
+    ) -> Result<()> {
+        let cutoffs = Cutoffs::from_progress(progress_runs);
+
+        // Trim data to per-run cutoffs for fair comparison
+        let disk_trimmed = cutoffs.trim_runs(disk_runs);
+        let memory_trimmed = cutoffs.trim_dual_runs(memory_runs);
+        let io_trimmed = cutoffs.trim_dual_runs(io_runs);
+
+        if !disk_trimmed.is_empty() {
+            chart::generate(
+                chart::ChartConfig {
+                    output_path: &crate_path.join("disk.svg"),
+                    title: format!("{} — Disk Usage", crate_name),
+                    y_label: "Disk Usage".to_string(),
+                    y_format: chart::YAxisFormat::Bytes,
+                },
+                &disk_trimmed,
+            )?;
+        }
+
+        if !memory_trimmed.is_empty() {
+            chart::generate_dual(
+                chart::ChartConfig {
+                    output_path: &crate_path.join("memory.svg"),
+                    title: format!("{} — Memory", crate_name),
+                    y_label: "Memory".to_string(),
+                    y_format: chart::YAxisFormat::Bytes,
+                },
+                &memory_trimmed,
+                "(current)",
+                "(peak)",
+            )?;
+        }
+
+        if !progress_runs.is_empty() {
+            let progress_trimmed = cutoffs.trim_runs(progress_runs);
+            chart::generate(
+                chart::ChartConfig {
+                    output_path: &crate_path.join("progress.svg"),
+                    title: format!("{} — Progress", crate_name),
+                    y_label: "Progress".to_string(),
+                    y_format: chart::YAxisFormat::Number,
+                },
+                &progress_trimmed,
+            )?;
+        }
+
+        if !io_trimmed.is_empty() {
+            // I/O Read (primary column)
+            let io_read: Vec<_> = io_trimmed
+                .iter()
+                .map(|r| Run {
+                    id: r.id.clone(),
+                    data: r.primary.clone(),
+                })
+                .collect();
+            chart::generate(
+                chart::ChartConfig {
+                    output_path: &crate_path.join("io_read.svg"),
+                    title: format!("{} — I/O Read", crate_name),
+                    y_label: "Bytes Read".to_string(),
+                    y_format: chart::YAxisFormat::Bytes,
+                },
+                &io_read,
+            )?;
+
+            // I/O Write (secondary column)
+            let io_write: Vec<_> = io_trimmed
+                .iter()
+                .map(|r| Run {
+                    id: r.id.clone(),
+                    data: r.secondary.clone(),
+                })
+                .collect();
+            chart::generate(
+                chart::ChartConfig {
+                    output_path: &crate_path.join("io_write.svg"),
+                    title: format!("{} — I/O Write", crate_name),
+                    y_label: "Bytes Written".to_string(),
+                    y_format: chart::YAxisFormat::Bytes,
+                },
+                &io_write,
+            )?;
+        }
+
+        Ok(())
+    }
+
+    fn generate_individual_charts(
+        &self,
+        crate_path: &Path,
+        crate_name: &str,
+        disk_runs: &[Run],
+        memory_runs: &[DualRun],
+        progress_runs: &[Run],
+        io_runs: &[DualRun],
+    ) -> Result<()> {
+        for run in disk_runs {
+            let run_path = crate_path.join(&run.id);
+            chart::generate(
+                chart::ChartConfig {
+                    output_path: &run_path.join("disk.svg"),
+                    title: format!("{} — Disk Usage", crate_name),
+                    y_label: "Disk Usage".to_string(),
+                    y_format: chart::YAxisFormat::Bytes,
+                },
+                std::slice::from_ref(run),
+            )?;
+        }
+
+        for run in memory_runs {
+            let run_path = crate_path.join(&run.id);
+            chart::generate_dual(
+                chart::ChartConfig {
+                    output_path: &run_path.join("memory.svg"),
+                    title: format!("{} — Memory", crate_name),
+                    y_label: "Memory".to_string(),
+                    y_format: chart::YAxisFormat::Bytes,
+                },
+                std::slice::from_ref(run),
+                "(current)",
+                "(peak)",
+            )?;
+        }
+
+        for run in progress_runs {
+            let run_path = crate_path.join(&run.id);
+            chart::generate(
+                chart::ChartConfig {
+                    output_path: &run_path.join("progress.svg"),
+                    title: format!("{} — Progress", crate_name),
+                    y_label: "Progress".to_string(),
+                    y_format: chart::YAxisFormat::Number,
+                },
+                std::slice::from_ref(run),
+            )?;
+        }
+
+        for run in io_runs {
+            let run_path = crate_path.join(&run.id);
+
+            let read_run = Run {
+                id: run.id.clone(),
+                data: run.primary.clone(),
+            };
+            chart::generate(
+                chart::ChartConfig {
+                    output_path: &run_path.join("io_read.svg"),
+                    title: format!("{} — I/O Read", crate_name),
+                    y_label: "Bytes Read".to_string(),
+                    y_format: chart::YAxisFormat::Bytes,
+                },
+                std::slice::from_ref(&read_run),
+            )?;
+
+            let write_run = Run {
+                id: run.id.clone(),
+                data: run.secondary.clone(),
+            };
+            chart::generate(
+                chart::ChartConfig {
+                    output_path: &run_path.join("io_write.svg"),
+                    title: format!("{} — I/O Write", crate_name),
+                    y_label: "Bytes Written".to_string(),
+                    y_format: chart::YAxisFormat::Bytes,
+                },
+                std::slice::from_ref(&write_run),
+            )?;
+        }
+
+        Ok(())
+    }
+}

crates/brk_bencher_visualizer/src/main.rs

@@ -0,0 +1,6 @@
+use brk_bencher_visualizer::Visualizer;
+
+fn main() {
+    let v = Visualizer::from_cargo_env().unwrap();
+    v.generate_all_charts().unwrap();
+}

crates/brk_bindgen/.gitignore

@@ -0,0 +1,6 @@
+clients/
+/*.json
+/*.js
+/*.rs
+/*.py
+tests/output/

crates/brk_bindgen/Cargo.toml

@@ -0,0 +1,17 @@
+[package]
+name = "brk_bindgen"
+description = "A trait-based generator of client bindings for multiple languages"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+homepage.workspace = true
+repository.workspace = true
+
+[dependencies]
+brk_cohort = { workspace = true }
+brk_query = { workspace = true }
+brk_types = { workspace = true }
+indexmap = { workspace = true }
+oas3 = "0.22"
+serde = { workspace = true }
+serde_json = { workspace = true }

crates/brk_bindgen/README.md

@@ -0,0 +1,46 @@
+# brk_bindgen
+
+Code generation for BRK client libraries.
+
+## What It Enables
+
+Generate typed client libraries for Rust, JavaScript, and Python from the OpenAPI specification. Keeps frontend code in sync with available metrics and API endpoints without manual maintenance.
+
+## Key Features
+
+- **Multi-language**: Generates Rust, JavaScript, and Python clients
+- **OpenAPI-driven**: Extracts endpoints and schemas from the OpenAPI spec
+- **Metric catalog**: Includes all metric IDs and their supported indexes
+- **Type definitions**: Generates types/interfaces from JSON Schema
+- **Selective output**: Generate only the languages you need
+
+## Core API
+
+```rust,ignore
+use brk_bindgen::{generate_clients, ClientOutputPaths};
+
+let paths = ClientOutputPaths::new()
+    .rust("crates/brk_client/src/lib.rs")
+    .javascript("modules/brk-client/index.js")
+    .python("packages/brk_client/brk_client/__init__.py");
+
+generate_clients(&vecs, &openapi_json, &paths)?;
+```
+
+## Generated Clients
+
+| Language | Contents |
+|----------|----------|
+| Rust | Typed API client using `brk_types`, metric catalog |
+| JavaScript | ES module with JSDoc types, metric catalog, fetch helpers |
+| Python | Typed client with dataclasses, metric catalog |
+
+Each client includes:
+- All REST API endpoints as typed functions
+- Complete metric catalog with index information
+- Type definitions for request/response schemas
+
+## Built On
+
+- `brk_query` for metric enumeration
+- `brk_types` for type schemas

crates/brk_bindgen/src/analysis/mod.rs

@@ -0,0 +1,14 @@
+//! Analysis module for name deconstruction and pattern detection.
+//!
+//! This module implements bottom-up analysis of vec names to detect
+//! common denominators (prefixes/suffixes) and field positions.
+
+mod names;
+mod patterns;
+mod positions;
+mod tree;
+
+pub use names::*;
+pub use patterns::*;
+pub use positions::*;
+pub use tree::*;

crates/brk_bindgen/src/analysis/names.rs

@@ -0,0 +1,198 @@
+//! Common prefix/suffix detection for series names.
+//!
+//! This module provides utilities to find common prefixes and suffixes
+//! among series names, which is used to detect pattern mode (suffix vs prefix).
+
+/// Find the longest common prefix among all strings.
+/// Returns the prefix WITH trailing underscore if found at word boundary.
+/// Returns None if no common prefix exists.
+pub fn find_common_prefix(names: &[&str]) -> Option<String> {
+    if names.is_empty() || names.iter().any(|n| n.is_empty()) {
+        return None;
+    }
+
+    let first = names[0];
+
+    // Find character-by-character common prefix
+    let mut prefix_len = 0;
+    for (i, ch) in first.chars().enumerate() {
+        if names.iter().all(|n| n.chars().nth(i) == Some(ch)) {
+            prefix_len = i + 1;
+        } else {
+            break;
+        }
+    }
+
+    if prefix_len == 0 {
+        return None;
+    }
+
+    let raw_prefix = &first[..prefix_len];
+
+    // Must end at underscore boundary for semantic coherence
+    if raw_prefix.ends_with('_') {
+        return Some(raw_prefix.to_string());
+    }
+
+    // If raw_prefix equals one of the full names (one name is a prefix of all others),
+    // return it with trailing underscore for proper base detection
+    if names.contains(&raw_prefix) {
+        return Some(format!("{}_", raw_prefix));
+    }
+
+    // Find the last underscore position
+    if let Some(last_underscore) = raw_prefix.rfind('_') {
+        let clean_prefix = &first[..=last_underscore];
+        if names.iter().all(|n| n.starts_with(clean_prefix)) {
+            return Some(clean_prefix.to_string());
+        }
+    }
+
+    None
+}
+
+/// Find the longest common suffix among all strings.
+/// Returns the suffix WITH leading underscore if found at word boundary.
+/// Returns None if no common suffix exists.
+pub fn find_common_suffix(names: &[&str]) -> Option<String> {
+    if names.is_empty() || names.iter().any(|n| n.is_empty()) {
+        return None;
+    }
+
+    let first = names[0];
+    let first_chars: Vec<char> = first.chars().collect();
+
+    // Find character-by-character common suffix (from the end)
+    let mut suffix_len = 0;
+    for i in 0..first_chars.len() {
+        let idx_from_end = first_chars.len() - 1 - i;
+        let ch = first_chars[idx_from_end];
+
+        let all_match = names.iter().all(|n| {
+            let n_chars: Vec<char> = n.chars().collect();
+            if i >= n_chars.len() {
+                return false;
+            }
+            n_chars[n_chars.len() - 1 - i] == ch
+        });
+
+        if all_match {
+            suffix_len = i + 1;
+        } else {
+            break;
+        }
+    }
+
+    if suffix_len == 0 {
+        return None;
+    }
+
+    let raw_suffix = &first[first.len() - suffix_len..];
+
+    // Must start at underscore boundary for semantic coherence
+    if raw_suffix.starts_with('_') {
+        return Some(raw_suffix.to_string());
+    }
+
+    // Check if preceded by underscore in all names (word boundary)
+    let at_word_boundary = names.iter().all(|n| {
+        if *n == raw_suffix {
+            true // Suffix is the whole string
+        } else if let Some(prefix) = n.strip_suffix(raw_suffix) {
+            prefix.ends_with('_')
+        } else {
+            false
+        }
+    });
+
+    if at_word_boundary {
+        return Some(format!("_{}", raw_suffix));
+    }
+
+    // Find the first underscore position in suffix
+    if let Some(first_underscore) = raw_suffix.find('_') {
+        let clean_suffix = &raw_suffix[first_underscore..];
+        if names.iter().all(|n| n.ends_with(clean_suffix)) {
+            return Some(clean_suffix.to_string());
+        }
+    }
+
+    None
+}
+
+/// Normalize a prefix string by ensuring it ends with underscore.
+/// Returns empty string if input is empty.
+pub fn normalize_prefix(s: &str) -> String {
+    if s.is_empty() {
+        String::new()
+    } else if s.ends_with('_') {
+        s.to_string()
+    } else {
+        format!("{}_", s)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_common_prefix_basic() {
+        let names = vec!["addrs_0sats", "addrs_1sats", "addrs_2sats"];
+        assert_eq!(find_common_prefix(&names), Some("addrs_".to_string()));
+    }
+
+    #[test]
+    fn test_common_prefix_none() {
+        let names = vec!["foo", "bar", "baz"];
+        assert_eq!(find_common_prefix(&names), None);
+    }
+
+    #[test]
+    fn test_common_prefix_lth() {
+        let names = vec!["lth_cost_basis_max", "lth_cost_basis_min", "lth_cost_basis"];
+        assert_eq!(
+            find_common_prefix(&names),
+            Some("lth_cost_basis_".to_string())
+        );
+    }
+
+    #[test]
+    fn test_common_suffix_basic() {
+        let names = vec!["cumulative_supply", "net_supply", "total_supply"];
+        assert_eq!(find_common_suffix(&names), Some("_supply".to_string()));
+    }
+
+    #[test]
+    fn test_common_prefix_cost_basis() {
+        // With suffix naming convention, cost_basis variants share a common prefix
+        let names = vec!["cost_basis_max", "cost_basis_min", "cost_basis"];
+        assert_eq!(find_common_prefix(&names), Some("cost_basis_".to_string()));
+    }
+
+    #[test]
+    fn test_common_suffix_none() {
+        let names = vec!["foo", "bar", "baz"];
+        assert_eq!(find_common_suffix(&names), None);
+    }
+
+    #[test]
+    fn test_common_prefix_one_is_prefix_of_other() {
+        // When one name is a prefix of another (block_count vs block_count_cumulative)
+        let names = vec!["block_count_cumulative", "block_count"];
+        assert_eq!(find_common_prefix(&names), Some("block_count_".to_string()));
+    }
+
+    #[test]
+    fn test_common_suffix_realized_loss() {
+        let names = vec![
+            "cumulative_realized_loss",
+            "net_realized_loss",
+            "realized_loss",
+        ];
+        assert_eq!(
+            find_common_suffix(&names),
+            Some("_realized_loss".to_string())
+        );
+    }
+}

crates/brk_bindgen/src/analysis/patterns.rs

@@ -0,0 +1,382 @@
+//! Structural pattern detection using bottom-up analysis.
+//!
+//! This module detects repeating tree structures and analyzes them
+//! using the bottom-up name deconstruction algorithm.
+
+use std::{
+    cmp::Reverse,
+    collections::{BTreeMap, BTreeSet},
+};
+
+use brk_types::{TreeNode, extract_json_type};
+
+use super::analyze_pattern_modes;
+use crate::{PatternBaseResult, PatternField, StructuralPattern, to_pascal_case};
+
+/// Context for pattern detection, holding all intermediate state.
+struct PatternContext {
+    /// Maps field signatures to pattern names
+    signature_to_pattern: BTreeMap<Vec<PatternField>, String>,
+    /// Counts how many times each signature appears
+    signature_counts: BTreeMap<Vec<PatternField>, usize>,
+    /// Maps normalized signatures to pattern names (for naming consistency)
+    normalized_to_name: BTreeMap<Vec<PatternField>, String>,
+    /// Counts pattern name usage (for unique naming)
+    name_counts: BTreeMap<String, usize>,
+    /// Maps signatures to their child field lists
+    signature_to_child_fields: BTreeMap<Vec<PatternField>, Vec<Vec<PatternField>>>,
+}
+
+impl PatternContext {
+    fn new() -> Self {
+        Self {
+            signature_to_pattern: BTreeMap::new(),
+            signature_counts: BTreeMap::new(),
+            normalized_to_name: BTreeMap::new(),
+            name_counts: BTreeMap::new(),
+            signature_to_child_fields: BTreeMap::new(),
+        }
+    }
+}
+
+/// Detect structural patterns in the tree using a bottom-up approach.
+///
+/// Returns (patterns, concrete_to_pattern, concrete_to_type_param, node_bases).
+/// Each pattern has its `mode` set based on analysis of all instances.
+/// `node_bases` maps tree paths to their computed PatternBaseResult for use during generation.
+pub fn detect_structural_patterns(
+    tree: &TreeNode,
+) -> (
+    Vec<StructuralPattern>,
+    BTreeMap<Vec<PatternField>, String>,
+    BTreeMap<Vec<PatternField>, String>,
+    BTreeMap<String, PatternBaseResult>,
+) {
+    let mut ctx = PatternContext::new();
+    resolve_branch_patterns(tree, &mut ctx);
+
+    let (generic_patterns, generic_mappings, type_mappings) =
+        detect_generic_patterns(&ctx.signature_to_pattern);
+
+    // Only include patterns that appear 2+ times for the patterns list
+    let mut patterns: Vec<StructuralPattern> = ctx
+        .signature_to_pattern
+        .iter()
+        .filter(|(sig, _)| {
+            ctx.signature_counts.get(*sig).copied().unwrap_or(0) >= 2
+                && !generic_mappings.contains_key(*sig)
+        })
+        .map(|(fields, name)| {
+            let child_fields_list = ctx.signature_to_child_fields.get(fields);
+            let fields_with_type_params = fields
+                .iter()
+                .enumerate()
+                .map(|(i, f)| {
+                    let type_param = child_fields_list
+                        .and_then(|list| list.get(i))
+                        .and_then(|cf| type_mappings.get(cf).cloned());
+                    PatternField {
+                        type_param,
+                        ..f.clone()
+                    }
+                })
+                .collect();
+            StructuralPattern {
+                name: name.clone(),
+                fields: fields_with_type_params,
+                mode: None, // Will be determined by analyze_pattern_modes
+                is_generic: false,
+            }
+        })
+        .collect();
+
+    // Deduplicate patterns by name - different signatures can map to the same name
+    // when their normalized forms match but they can't be unified as generics
+    {
+        let mut seen_names: BTreeSet<String> = BTreeSet::new();
+        patterns.retain(|p| seen_names.insert(p.name.clone()));
+    }
+
+    patterns.extend(generic_patterns);
+
+    // Build pattern lookup for mode analysis (patterns appearing 2+ times)
+    let mut pattern_lookup: BTreeMap<Vec<PatternField>, String> = BTreeMap::new();
+    for (sig, name) in &ctx.signature_to_pattern {
+        if ctx.signature_counts.get(sig).copied().unwrap_or(0) >= 2 {
+            pattern_lookup.insert(sig.clone(), name.clone());
+        }
+    }
+    pattern_lookup.extend(generic_mappings.clone());
+
+    let concrete_to_pattern = pattern_lookup.clone();
+
+    // Analyze pattern modes (suffix vs prefix) from all instances
+    // Also collects node bases for each tree path
+    let node_bases = analyze_pattern_modes(tree, &mut patterns, &pattern_lookup);
+
+    patterns.sort_by_key(|p| Reverse(p.fields.len()));
+    (patterns, concrete_to_pattern, type_mappings, node_bases)
+}
+
+/// Detect generic patterns by grouping signatures by their normalized form.
+fn detect_generic_patterns(
+    signature_to_pattern: &BTreeMap<Vec<PatternField>, String>,
+) -> (
+    Vec<StructuralPattern>,
+    BTreeMap<Vec<PatternField>, String>,
+    BTreeMap<Vec<PatternField>, String>,
+) {
+    let mut normalized_groups: BTreeMap<
+        Vec<PatternField>,
+        Vec<(Vec<PatternField>, String, String)>,
+    > = BTreeMap::new();
+
+    for (fields, name) in signature_to_pattern {
+        if let Some((normalized, extracted_type)) = normalize_fields_for_generic(fields) {
+            normalized_groups.entry(normalized).or_default().push((
+                fields.clone(),
+                name.clone(),
+                extracted_type,
+            ));
+        }
+    }
+
+    let mut patterns = Vec::new();
+    let mut pattern_mappings: BTreeMap<Vec<PatternField>, String> = BTreeMap::new();
+    let mut type_mappings: BTreeMap<Vec<PatternField>, String> = BTreeMap::new();
+
+    for (normalized_fields, group) in normalized_groups {
+        if group.len() >= 2 {
+            let generic_name = group[0].1.clone();
+            for (concrete_fields, _, extracted_type) in &group {
+                pattern_mappings.insert(concrete_fields.clone(), generic_name.clone());
+                type_mappings.insert(concrete_fields.clone(), extracted_type.clone());
+            }
+            patterns.push(StructuralPattern {
+                name: generic_name,
+                fields: normalized_fields,
+                mode: None, // Will be determined by analyze_pattern_modes
+                is_generic: true,
+            });
+        }
+    }
+
+    (patterns, pattern_mappings, type_mappings)
+}
+
+/// Normalize fields by replacing concrete value types with "T".
+///
+/// Handles two cases:
+/// 1. All leaves have identical types (e.g., all `Sats`) -> normalize to `T`
+/// 2. All leaves have wrapper types with the same inner type (e.g., `Open<Sats>`, `High<Sats>`)
+///    -> normalize to `Open<T>`, `High<T>`, etc.
+fn normalize_fields_for_generic(fields: &[PatternField]) -> Option<(Vec<PatternField>, String)> {
+    let leaf_types: Vec<&str> = fields
+        .iter()
+        .filter(|f| f.is_leaf())
+        .map(|f| f.rust_type.as_str())
+        .collect();
+
+    if leaf_types.is_empty() {
+        return None;
+    }
+
+    let first_type = leaf_types[0];
+
+    // Case 1: All leaf types are identical
+    if leaf_types.iter().all(|t| *t == first_type) {
+        let normalized = fields
+            .iter()
+            .map(|f| {
+                if f.is_branch() {
+                    f.clone()
+                } else {
+                    PatternField {
+                        name: f.name.clone(),
+                        rust_type: "T".to_string(),
+                        json_type: "T".to_string(),
+                        indexes: f.indexes.clone(),
+                        type_param: None,
+                    }
+                }
+            })
+            .collect();
+        return Some((normalized, crate::extract_inner_type(first_type)));
+    }
+
+    // Case 2: Check if all leaves have wrapper types with the same inner type
+    // e.g., Open<Sats>, High<Sats>, Low<Sats>, Close<Sats> all have inner type Sats
+    let inner_types: Vec<String> = leaf_types
+        .iter()
+        .map(|t| crate::extract_inner_type(t))
+        .collect();
+
+    let first_inner = &inner_types[0];
+
+    // Only proceed if inner types differ from originals (meaning they had wrappers)
+    // and all inner types are the same
+    if inner_types.iter().all(|t| t == first_inner)
+        && inner_types
+            .iter()
+            .zip(leaf_types.iter())
+            .any(|(inner, orig)| inner != *orig)
+    {
+        let normalized = fields
+            .iter()
+            .map(|f| {
+                if f.is_branch() {
+                    f.clone()
+                } else {
+                    PatternField {
+                        name: f.name.clone(),
+                        rust_type: replace_inner_type(&f.rust_type, "T"),
+                        json_type: replace_inner_type(&f.json_type, "T"),
+                        indexes: f.indexes.clone(),
+                        type_param: None,
+                    }
+                }
+            })
+            .collect();
+        return Some((normalized, first_inner.clone()));
+    }
+
+    None
+}
+
+/// Replace the inner type of a wrapper generic with a new type.
+/// e.g., `Open<Sats>` with replacement `T` -> `Open<T>`
+fn replace_inner_type(type_str: &str, replacement: &str) -> String {
+    if let Some(start) = type_str.find('<')
+        && let Some(end) = type_str.rfind('>')
+        && start < end
+    {
+        format!("{}<{}>", &type_str[..start], replacement)
+    } else {
+        replacement.to_string()
+    }
+}
+
+/// Recursively resolve branch patterns bottom-up.
+fn resolve_branch_patterns(
+    node: &TreeNode,
+    ctx: &mut PatternContext,
+) -> Option<(String, Vec<PatternField>)> {
+    let TreeNode::Branch(children) = node else {
+        return None;
+    };
+
+    // Convert to sorted BTreeMap for consistent pattern detection
+    let sorted_children: BTreeMap<_, _> = children.iter().collect();
+
+    let mut fields: Vec<PatternField> = Vec::new();
+    let mut child_fields_vec: Vec<Vec<PatternField>> = Vec::new();
+
+    for (child_name, child_node) in sorted_children {
+        let (rust_type, json_type, indexes, child_fields) = match child_node {
+            TreeNode::Leaf(leaf) => (
+                leaf.kind().to_string(),
+                extract_json_type(&leaf.schema),
+                leaf.indexes().clone(),
+                Vec::new(),
+            ),
+            TreeNode::Branch(_) => {
+                let (pattern_name, child_pattern_fields) = resolve_branch_patterns(child_node, ctx)
+                    .unwrap_or_else(|| ("Unknown".to_string(), Vec::new()));
+                (
+                    pattern_name.clone(),
+                    pattern_name,
+                    BTreeSet::new(),
+                    child_pattern_fields,
+                )
+            }
+        };
+        fields.push(PatternField {
+            name: child_name.clone(),
+            rust_type,
+            json_type,
+            indexes,
+            type_param: None,
+        });
+        child_fields_vec.push(child_fields);
+    }
+
+    // Fields are already sorted since we iterated over BTreeMap
+    *ctx.signature_counts.entry(fields.clone()).or_insert(0) += 1;
+
+    ctx.signature_to_child_fields
+        .entry(fields.clone())
+        .or_insert(child_fields_vec);
+
+    let pattern_name = if let Some(existing) = ctx.signature_to_pattern.get(&fields) {
+        existing.clone()
+    } else {
+        let normalized = normalize_fields_for_naming(&fields);
+        // Generate stable name from first word of each field (deduped, sorted)
+        let first_words: BTreeSet<String> = fields
+            .iter()
+            .filter_map(|f| f.name.split('_').next())
+            .map(to_pascal_case)
+            .collect();
+        let combined: String = first_words.into_iter().collect();
+        let name = ctx
+            .normalized_to_name
+            .entry(normalized)
+            .or_insert_with(|| generate_pattern_name(&combined, &mut ctx.name_counts))
+            .clone();
+        ctx.signature_to_pattern
+            .insert(fields.clone(), name.clone());
+        name
+    };
+
+    Some((pattern_name, fields))
+}
+
+/// Normalize fields for naming (same structure = same name).
+/// Only erases leaf types when all leaves share the same type — this ensures
+/// mixed-type signatures (e.g., StoredU32 raw + StoredU64 cumulative) get a
+/// different name than same-type signatures that can be genericized.
+fn normalize_fields_for_naming(fields: &[PatternField]) -> Vec<PatternField> {
+    let leaf_types: Vec<&str> = fields
+        .iter()
+        .filter(|f| !f.is_branch())
+        .map(|f| f.rust_type.as_str())
+        .collect();
+    let all_same = !leaf_types.is_empty() && leaf_types.iter().all(|t| *t == leaf_types[0]);
+
+    fields
+        .iter()
+        .map(|f| {
+            if f.is_branch() || !all_same {
+                f.clone()
+            } else {
+                PatternField {
+                    name: f.name.clone(),
+                    rust_type: "_".to_string(),
+                    json_type: "_".to_string(),
+                    indexes: f.indexes.clone(),
+                    type_param: None,
+                }
+            }
+        })
+        .collect()
+}
+
+/// Generate a unique pattern name.
+fn generate_pattern_name(field_name: &str, name_counts: &mut BTreeMap<String, usize>) -> String {
+    let pascal = to_pascal_case(field_name);
+    let sanitized = if pascal.chars().next().is_some_and(|c| c.is_ascii_digit()) {
+        format!("_{}", pascal)
+    } else {
+        pascal
+    };
+
+    let base_name = format!("{}Pattern", sanitized);
+    let count = name_counts.entry(base_name.clone()).or_insert(0);
+    *count += 1;
+
+    if *count == 1 {
+        base_name
+    } else {
+        format!("{}{}", base_name, count)
+    }
+}

crates/brk_bindgen/src/analysis/tree.rs

@@ -0,0 +1,573 @@
+//! Tree traversal helpers for pattern analysis.
+//!
+//! This module provides utilities for working with the TreeNode structure,
+//! including leaf name extraction and index pattern detection.
+
+use std::collections::{BTreeMap, BTreeSet};
+
+use brk_types::{Index, TreeNode, extract_json_type};
+use indexmap::IndexMap;
+
+use crate::{IndexSetPattern, PatternField, child_type_name};
+
+use super::{find_common_prefix, find_common_suffix, normalize_prefix};
+
+/// Get the shortest leaf name from a tree node.
+///
+/// This is useful for pattern base analysis where we want the "base" case
+/// (e.g., the leaf without suffix like `_btc` or `_usd`).
+pub(super) fn get_shortest_leaf_name(node: &TreeNode) -> Option<String> {
+    match node {
+        TreeNode::Leaf(leaf) => Some(leaf.name().to_string()),
+        TreeNode::Branch(children) => children
+            .values()
+            .filter_map(get_shortest_leaf_name)
+            .min_by_key(|name| name.len()),
+    }
+}
+
+/// Get the field signature for a branch node's children.
+/// Fields are sorted alphabetically for consistent pattern matching.
+pub fn get_node_fields(
+    children: &IndexMap<String, TreeNode>,
+    pattern_lookup: &BTreeMap<Vec<PatternField>, String>,
+) -> Vec<PatternField> {
+    let mut fields: Vec<PatternField> = children
+        .iter()
+        .map(|(name, node)| {
+            let (rust_type, json_type, indexes) = match node {
+                TreeNode::Leaf(leaf) => (
+                    leaf.kind().to_string(),
+                    extract_json_type(&leaf.schema),
+                    leaf.indexes().clone(),
+                ),
+                TreeNode::Branch(grandchildren) => {
+                    let child_fields = get_node_fields(grandchildren, pattern_lookup);
+                    let pattern_name = pattern_lookup
+                        .get(&child_fields)
+                        .cloned()
+                        .unwrap_or_else(|| "Unknown".to_string());
+                    (pattern_name.clone(), pattern_name, BTreeSet::new())
+                }
+            };
+            PatternField {
+                name: name.clone(),
+                rust_type,
+                json_type,
+                indexes,
+                type_param: None,
+            }
+        })
+        .collect();
+    // Sort for consistent pattern matching (display order preserved in IndexMap)
+    fields.sort_by(|a, b| a.name.cmp(&b.name));
+    fields
+}
+
+/// Detect index patterns (sets of indexes that appear together on series).
+pub fn detect_index_patterns(tree: &TreeNode) -> Vec<IndexSetPattern> {
+    let mut unique_index_sets: BTreeSet<BTreeSet<Index>> = BTreeSet::new();
+    collect_index_sets_from_tree(tree, &mut unique_index_sets);
+
+    // Sort by count (descending) then by first index name for deterministic ordering
+    let mut sorted_sets: Vec<_> = unique_index_sets
+        .into_iter()
+        .filter(|indexes| !indexes.is_empty())
+        .collect();
+    sorted_sets.sort_by(|a, b| {
+        b.len()
+            .cmp(&a.len())
+            .then_with(|| a.iter().next().cmp(&b.iter().next()))
+    });
+
+    // Assign unique sequential names
+    sorted_sets
+        .into_iter()
+        .enumerate()
+        .map(|(i, indexes)| IndexSetPattern {
+            name: format!("SeriesPattern{}", i + 1),
+            indexes,
+        })
+        .collect()
+}
+
+fn collect_index_sets_from_tree(
+    node: &TreeNode,
+    unique_index_sets: &mut BTreeSet<BTreeSet<Index>>,
+) {
+    match node {
+        TreeNode::Leaf(leaf) => {
+            unique_index_sets.insert(leaf.indexes().clone());
+        }
+        TreeNode::Branch(children) => {
+            for child in children.values() {
+                collect_index_sets_from_tree(child, unique_index_sets);
+            }
+        }
+    }
+}
+
+/// Result of analyzing a pattern instance's base.
+#[derive(Debug, Clone)]
+pub struct PatternBaseResult {
+    /// The computed base name for the pattern.
+    pub base: String,
+    /// Whether an outlier child was excluded to find the pattern.
+    /// If true, pattern factory should not be used.
+    pub has_outlier: bool,
+    /// Whether this instance uses suffix mode (common prefix) or prefix mode (common suffix).
+    /// Used to check compatibility with the pattern's mode.
+    pub is_suffix_mode: bool,
+    /// The field parts (suffix in suffix mode, prefix in prefix mode) for each field.
+    /// Used to check if instance field parts match the pattern's field parts.
+    pub field_parts: BTreeMap<String, String>,
+}
+
+impl PatternBaseResult {
+    /// Create a default result that forces inlining (has_outlier = true).
+    /// Use when no pattern base could be computed during lookup.
+    pub fn force_inline() -> Self {
+        Self {
+            base: String::new(),
+            has_outlier: true,
+            is_suffix_mode: true,
+            field_parts: BTreeMap::new(),
+        }
+    }
+
+    /// Create an empty result with no outlier.
+    /// Use for root-level patterns or when children have no common pattern.
+    pub fn empty() -> Self {
+        Self {
+            base: String::new(),
+            has_outlier: false,
+            is_suffix_mode: true,
+            field_parts: BTreeMap::new(),
+        }
+    }
+}
+
+/// Get the series base for a pattern instance by analyzing direct children.
+///
+/// Uses the shortest leaf names from direct children to find common prefix/suffix.
+///
+/// If the initial analysis fails to find a common pattern, it tries excluding
+/// each child one at a time to detect outliers (e.g., a mismatched "base" field
+/// from indexer/computed tree merging).
+///
+/// Returns both the base and whether an outlier was detected.
+pub fn get_pattern_instance_base(node: &TreeNode) -> PatternBaseResult {
+    let child_names = get_direct_children_for_analysis(node);
+    if child_names.is_empty() {
+        return PatternBaseResult::empty();
+    }
+
+    // Try to find common base from leaf names
+    if let Some(result) = try_find_base(&child_names, false) {
+        return PatternBaseResult {
+            base: result.base,
+            has_outlier: result.has_outlier,
+            is_suffix_mode: result.is_suffix_mode,
+            field_parts: result.field_parts,
+        };
+    }
+
+    // If no common pattern found and we have enough children, try excluding outliers
+    if child_names.len() > 2 {
+        for i in 0..child_names.len() {
+            let filtered: Vec<_> = child_names
+                .iter()
+                .enumerate()
+                .filter(|(j, _)| *j != i)
+                .map(|(_, v)| v.clone())
+                .collect();
+
+            if let Some(result) = try_find_base(&filtered, true) {
+                return PatternBaseResult {
+                    base: result.base,
+                    has_outlier: true,
+                    is_suffix_mode: result.is_suffix_mode,
+                    field_parts: result.field_parts,
+                };
+            }
+        }
+    }
+
+    // Fallback: no common prefix/suffix found - this is a root-level pattern
+    // Return empty base so series names are used directly
+    PatternBaseResult::empty()
+}
+
+/// Result of try_find_base: base name, has_outlier flag, is_suffix_mode flag, and field_parts.
+struct FindBaseResult {
+    base: String,
+    has_outlier: bool,
+    is_suffix_mode: bool,
+    field_parts: BTreeMap<String, String>,
+}
+
+/// Try to find a common base from child names using prefix/suffix detection.
+/// Returns Some(FindBaseResult) if found.
+fn try_find_base(
+    child_names: &[(String, String)],
+    is_outlier_attempt: bool,
+) -> Option<FindBaseResult> {
+    let leaf_names: Vec<&str> = child_names.iter().map(|(_, n)| n.as_str()).collect();
+
+    // Try common prefix first (suffix mode)
+    if let Some(prefix) = find_common_prefix(&leaf_names) {
+        let base = prefix.trim_end_matches('_').to_string();
+        let mut field_parts = BTreeMap::new();
+        for (field_name, leaf_name) in child_names {
+            // Compute the suffix part for this field
+            let suffix = if leaf_name == &base {
+                String::new()
+            } else {
+                leaf_name
+                    .strip_prefix(&prefix)
+                    .unwrap_or(leaf_name)
+                    .to_string()
+            };
+            field_parts.insert(field_name.clone(), suffix);
+        }
+        return Some(FindBaseResult {
+            base,
+            has_outlier: is_outlier_attempt,
+            is_suffix_mode: true,
+            field_parts,
+        });
+    }
+
+    // Try common suffix (prefix mode)
+    if let Some(suffix) = find_common_suffix(&leaf_names) {
+        let base = suffix.trim_start_matches('_').to_string();
+        let mut field_parts = BTreeMap::new();
+        for (field_name, leaf_name) in child_names {
+            // Compute the prefix part for this field, normalized to end with _
+            let prefix_part = leaf_name
+                .strip_suffix(&suffix)
+                .map(normalize_prefix)
+                .unwrap_or_default();
+            field_parts.insert(field_name.clone(), prefix_part);
+        }
+        return Some(FindBaseResult {
+            base,
+            has_outlier: is_outlier_attempt,
+            is_suffix_mode: false,
+            field_parts,
+        });
+    }
+
+    None
+}
+
+/// Get (field_name, shortest_leaf_name) pairs for direct children of a branch node.
+///
+/// Uses the shortest leaf name from each child subtree to find the "base" case
+/// (the leaf without suffix modifiers like `_btc` or `_usd`).
+fn get_direct_children_for_analysis(node: &TreeNode) -> Vec<(String, String)> {
+    match node {
+        TreeNode::Leaf(leaf) => vec![(leaf.name().to_string(), leaf.name().to_string())],
+        TreeNode::Branch(children) => children
+            .iter()
+            .filter_map(|(field_name, child)| {
+                get_shortest_leaf_name(child).map(|leaf_name| (field_name.clone(), leaf_name))
+            })
+            .collect(),
+    }
+}
+
+/// Infer the accumulated name for a child node based on a descendant leaf name.
+pub fn infer_accumulated_name(parent_acc: &str, field_name: &str, descendant_leaf: &str) -> String {
+    if let Some(pos) = descendant_leaf.find(field_name) {
+        if pos == 0 {
+            return field_name.to_string();
+        }
+        if pos > 0 && descendant_leaf.chars().nth(pos - 1) == Some('_') {
+            return if parent_acc.is_empty() {
+                field_name.to_string()
+            } else {
+                format!("{}_{}", parent_acc, field_name)
+            };
+        }
+    }
+
+    if parent_acc.is_empty() {
+        field_name.to_string()
+    } else {
+        format!("{}_{}", parent_acc, field_name)
+    }
+}
+
+/// Get fields with child field information for generic pattern lookup.
+pub fn get_fields_with_child_info(
+    children: &IndexMap<String, TreeNode>,
+    parent_name: &str,
+    pattern_lookup: &BTreeMap<Vec<PatternField>, String>,
+) -> Vec<(PatternField, Option<Vec<PatternField>>)> {
+    children
+        .iter()
+        .map(|(name, node)| {
+            let (rust_type, json_type, indexes, child_fields) = match node {
+                TreeNode::Leaf(leaf) => (
+                    leaf.kind().to_string(),
+                    extract_json_type(&leaf.schema),
+                    leaf.indexes().clone(),
+                    None,
+                ),
+                TreeNode::Branch(grandchildren) => {
+                    let child_fields = get_node_fields(grandchildren, pattern_lookup);
+                    let pattern_name = pattern_lookup
+                        .get(&child_fields)
+                        .cloned()
+                        .unwrap_or_else(|| child_type_name(parent_name, name));
+                    (
+                        pattern_name.clone(),
+                        pattern_name,
+                        BTreeSet::new(),
+                        Some(child_fields),
+                    )
+                }
+            };
+            (
+                PatternField {
+                    name: name.clone(),
+                    rust_type,
+                    json_type,
+                    indexes,
+                    type_param: None,
+                },
+                child_fields,
+            )
+        })
+        .collect()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use brk_types::{SeriesLeaf, SeriesLeafWithSchema, TreeNode};
+
+    fn make_leaf(name: &str) -> TreeNode {
+        let leaf = SeriesLeaf {
+            name: name.to_string(),
+            kind: "TestType".to_string(),
+            indexes: BTreeSet::new(),
+        };
+        TreeNode::Leaf(SeriesLeafWithSchema::new(leaf, serde_json::json!({})))
+    }
+
+    fn make_branch(children: Vec<(&str, TreeNode)>) -> TreeNode {
+        let map: IndexMap<String, TreeNode> = children
+            .into_iter()
+            .map(|(k, v)| (k.to_string(), v))
+            .collect();
+        TreeNode::Branch(map)
+    }
+
+    #[test]
+    fn test_get_pattern_instance_base_with_base_field() {
+        // Simulates vbytes tree: has base field with block_vbytes leaf
+        let tree = make_branch(vec![
+            (
+                "base",
+                make_branch(vec![("day1", make_leaf("block_vbytes"))]),
+            ),
+            (
+                "average",
+                make_branch(vec![("day1", make_leaf("block_vbytes_average"))]),
+            ),
+            (
+                "sum",
+                make_branch(vec![("day1", make_leaf("block_vbytes_sum"))]),
+            ),
+        ]);
+
+        let result = get_pattern_instance_base(&tree);
+        assert_eq!(result.base, "block_vbytes");
+        assert!(!result.has_outlier);
+    }
+
+    #[test]
+    fn test_get_pattern_instance_base_without_base_field() {
+        // Simulates weight tree: NO base field, only suffixed series
+        let tree = make_branch(vec![
+            (
+                "average",
+                make_branch(vec![("day1", make_leaf("block_weight_average"))]),
+            ),
+            (
+                "sum",
+                make_branch(vec![("day1", make_leaf("block_weight_sum"))]),
+            ),
+            (
+                "cumulative",
+                make_branch(vec![("day1", make_leaf("block_weight_cumulative"))]),
+            ),
+            (
+                "max",
+                make_branch(vec![("day1", make_leaf("block_weight_max"))]),
+            ),
+            (
+                "min",
+                make_branch(vec![("day1", make_leaf("block_weight_min"))]),
+            ),
+        ]);
+
+        let result = get_pattern_instance_base(&tree);
+        assert_eq!(result.base, "block_weight");
+        assert!(!result.has_outlier);
+    }
+
+    #[test]
+    fn test_get_pattern_instance_base_with_duplicate_base_field() {
+        // What if there's a "base" field that points to the same leaf as "average"?
+        // This could happen if the tree generation creates a base field that shares leaves with average
+        let tree = make_branch(vec![
+            (
+                "base",
+                make_branch(vec![("day1", make_leaf("block_weight_average"))]),
+            ),
+            (
+                "average",
+                make_branch(vec![("day1", make_leaf("block_weight_average"))]),
+            ),
+            (
+                "sum",
+                make_branch(vec![("day1", make_leaf("block_weight_sum"))]),
+            ),
+        ]);
+
+        let result = get_pattern_instance_base(&tree);
+        // Common prefix among all children is "block_weight_"
+        assert_eq!(result.base, "block_weight");
+        assert!(!result.has_outlier);
+    }
+
+    #[test]
+    fn test_get_pattern_instance_base_with_mismatched_base_name() {
+        // Simulates the actual bug: indexed tree's "base" field has name "weight"
+        // but computed tree's derived series use "block_weight_*" prefix.
+        // After tree merge, we get a base field with mismatched naming.
+        let tree = make_branch(vec![
+            ("base", make_leaf("weight")), // Outlier - doesn't match pattern
+            ("average", make_leaf("block_weight_average")),
+            ("sum", make_leaf("block_weight_sum")),
+            ("cumulative", make_leaf("block_weight_cumulative")),
+            ("max", make_leaf("block_weight_max")),
+            ("min", make_leaf("block_weight_min")),
+        ]);
+
+        let result = get_pattern_instance_base(&tree);
+        // Should detect "weight" as outlier and find common prefix from others
+        assert_eq!(result.base, "block_weight");
+        assert!(result.has_outlier); // Pattern factory should NOT be used
+    }
+
+    #[test]
+    fn test_get_pattern_instance_base_root_level_no_common_pattern() {
+        // Simulates root-level pattern with series that have no common prefix/suffix.
+        // These names have no shared prefix or suffix, even when excluding any one.
+        // In this case, we should return empty base so series names are used directly.
+        let tree = make_branch(vec![
+            ("alpha", make_leaf("foo_series")),
+            ("beta", make_leaf("bar_value")),
+            ("gamma", make_leaf("baz_count")),
+        ]);
+
+        let result = get_pattern_instance_base(&tree);
+        // No common prefix or suffix - return empty base
+        assert_eq!(result.base, "");
+        assert!(!result.has_outlier);
+    }
+
+    #[test]
+    fn test_get_pattern_instance_base_two_children_no_pattern() {
+        // Two children with no common pattern - should still return empty base
+        let tree = make_branch(vec![
+            ("foo", make_leaf("alpha")),
+            ("bar", make_leaf("beta")),
+        ]);
+
+        let result = get_pattern_instance_base(&tree);
+        assert_eq!(result.base, "");
+        assert!(!result.has_outlier);
+    }
+
+    #[test]
+    fn test_get_pattern_instance_base_with_outlier_excluded() {
+        // Simulates the realized pattern: adjusted_sopr, sopr, asopr.
+        // When "asopr" is excluded as outlier, "adjusted_sopr" and "sopr" share suffix "_sopr".
+        // The outlier detection should find base="sopr" with has_outlier=true.
+        let tree = make_branch(vec![
+            ("adjustedSopr", make_leaf("adjusted_sopr")),
+            ("sopr", make_leaf("sopr")),
+            ("asopr", make_leaf("asopr")),
+        ]);
+
+        let result = get_pattern_instance_base(&tree);
+        // Outlier detected - pattern base found by excluding "asopr"
+        assert_eq!(result.base, "sopr");
+        assert!(result.has_outlier); // Pattern factory should NOT be used (inline instead)
+    }
+
+    #[test]
+    fn test_get_pattern_instance_base_suffix_mode_price_ago() {
+        // Simulates price_ago pattern: price_24h_ago, price_1w_ago, price_10y_ago
+        // Common prefix is "price_", so this is suffix mode
+        let tree = make_branch(vec![
+            ("_24h", make_leaf("price_24h_ago")),
+            ("_1w", make_leaf("price_1w_ago")),
+            ("_1m", make_leaf("price_1m_ago")),
+            ("_10y", make_leaf("price_10y_ago")),
+        ]);
+
+        let result = get_pattern_instance_base(&tree);
+        assert_eq!(result.base, "price");
+        assert!(result.is_suffix_mode); // Suffix mode: _m(base, "24h_ago")
+        assert!(!result.has_outlier);
+    }
+
+    #[test]
+    fn test_get_pattern_instance_base_prefix_mode_price_returns() {
+        // Simulates price_returns pattern: 24h_price_returns, 1w_price_returns, 10y_price_returns
+        // Common suffix is "_price_returns", so this is prefix mode
+        let tree = make_branch(vec![
+            ("_24h", make_leaf("24h_price_returns")),
+            ("_1w", make_leaf("1w_price_returns")),
+            ("_1m", make_leaf("1m_price_returns")),
+            ("_10y", make_leaf("10y_price_returns")),
+        ]);
+
+        let result = get_pattern_instance_base(&tree);
+        assert_eq!(result.base, "price_returns");
+        assert!(!result.is_suffix_mode); // Prefix mode: _p("24h_", base)
+        assert!(!result.has_outlier);
+    }
+
+    #[test]
+    fn test_mode_detection_distinguishes_similar_structures() {
+        // Two patterns with identical structure but different naming conventions
+        // should have different modes detected
+
+        // Suffix mode pattern
+        let suffix_tree = make_branch(vec![
+            ("_1y", make_leaf("lump_sum_1y")),
+            ("_2y", make_leaf("lump_sum_2y")),
+            ("_5y", make_leaf("lump_sum_5y")),
+        ]);
+        let suffix_result = get_pattern_instance_base(&suffix_tree);
+        assert_eq!(suffix_result.base, "lump_sum");
+        assert!(suffix_result.is_suffix_mode);
+
+        // Prefix mode pattern (same structure, different naming)
+        let prefix_tree = make_branch(vec![
+            ("_1y", make_leaf("1y_returns")),
+            ("_2y", make_leaf("2y_returns")),
+            ("_5y", make_leaf("5y_returns")),
+        ]);
+        let prefix_result = get_pattern_instance_base(&prefix_tree);
+        assert_eq!(prefix_result.base, "returns");
+        assert!(!prefix_result.is_suffix_mode);
+    }
+}

crates/brk_bindgen/src/backends/javascript.rs

@@ -0,0 +1,109 @@
+//! JavaScript language syntax implementation.
+
+use crate::{GenericSyntax, LanguageSyntax, to_camel_case};
+
+/// JavaScript-specific code generation syntax.
+pub struct JavaScriptSyntax;
+
+impl LanguageSyntax for JavaScriptSyntax {
+    fn field_name(&self, name: &str) -> String {
+        to_camel_case(name)
+    }
+
+    fn path_expr(&self, base_var: &str, suffix: &str) -> String {
+        // Convert base_var to camelCase for JavaScript
+        let var_name = to_camel_case(base_var);
+        format!("`${{{}}}{}`", var_name, suffix)
+    }
+
+    fn suffix_expr(&self, acc_var: &str, relative: &str) -> String {
+        let var_name = to_camel_case(acc_var);
+        if relative.is_empty() {
+            // Identity: just return acc
+            var_name
+        } else {
+            // _m(acc, relative) -> acc ? `${acc}_relative` : 'relative'
+            format!("_m({}, '{}')", var_name, relative)
+        }
+    }
+
+    fn prefix_expr(&self, prefix: &str, acc_var: &str) -> String {
+        let var_name = to_camel_case(acc_var);
+        if prefix.is_empty() {
+            // Identity: just return acc
+            var_name
+        } else {
+            // _p(prefix, acc) -> acc ? `${prefix}${acc}` : 'prefix_without_underscore'
+            let prefix_base = prefix.trim_end_matches('_');
+            format!("_p('{}', {})", prefix_base, var_name)
+        }
+    }
+
+    fn constructor(&self, type_name: &str, path_expr: &str) -> String {
+        format!("create{}(client, {})", type_name, path_expr)
+    }
+
+    fn field_init(&self, indent: &str, name: &str, _type_ann: &str, value: &str) -> String {
+        // JavaScript uses object literal syntax; type is in JSDoc, not in assignment
+        format!("{}{}: {},", indent, name, value)
+    }
+
+    fn generic_syntax(&self) -> GenericSyntax {
+        GenericSyntax::JAVASCRIPT
+    }
+
+    fn string_literal(&self, value: &str) -> String {
+        format!("'{}'", value)
+    }
+
+    fn constructor_name(&self, type_name: &str) -> String {
+        format!("create{}", type_name)
+    }
+
+    fn disc_arg_expr(&self, template: &str) -> String {
+        if template == "{disc}" {
+            "disc".to_string()
+        } else if template.is_empty() {
+            "''".to_string()
+        } else if !template.contains("{disc}") {
+            format!("'{}'", template)
+        } else if template.ends_with("{disc}") {
+            let static_part = template.trim_end_matches("{disc}").trim_end_matches('_');
+            format!("_m('{}', disc)", static_part)
+        } else {
+            let js_template = template.replace("{disc}", "${disc}");
+            format!("`{}`", js_template)
+        }
+    }
+
+    fn template_expr(&self, acc_var: &str, template: &str) -> String {
+        let var_name = to_camel_case(acc_var);
+        if template.is_empty() {
+            // Identity — just pass disc
+            format!("_m({}, disc)", var_name)
+        } else if template == "{disc}" {
+            // Template IS the discriminator
+            format!("_m({}, disc)", var_name)
+        } else if !template.contains("{disc}") {
+            // Static suffix — no disc involved
+            format!("_m({}, '{}')", var_name, template)
+        } else {
+            // Template with {disc}: use nested _m for proper separator handling
+            // "ratio_{disc}_bps" → split on {disc} → _m(_m(acc, 'ratio'), disc) then _bps
+            // But this is complex. For embedded disc, use string interpolation.
+            // For suffix disc (ends with {disc}), use _m composition.
+            if let Some(static_part) = template.strip_suffix("{disc}") {
+                if static_part.is_empty() {
+                    format!("_m({}, disc)", var_name)
+                } else {
+                    let static_part = static_part.trim_end_matches('_');
+                    format!("_m(_m({}, '{}'), disc)", var_name, static_part)
+                }
+            } else {
+                // Embedded disc — use template literal
+                let js_template = template.replace("{disc}", "${disc}");
+                format!("_m({}, `{}`)", var_name, js_template)
+            }
+        }
+    }
+}

crates/brk_bindgen/src/backends/mod.rs

@@ -0,0 +1,12 @@
+//! Language-specific syntax backends.
+//!
+//! This module contains implementations of the `LanguageSyntax` trait
+//! for each supported target language.
+
+mod javascript;
+mod python;
+mod rust;
+
+pub use javascript::JavaScriptSyntax;
+pub use python::PythonSyntax;
+pub use rust::RustSyntax;

crates/brk_bindgen/src/backends/python.rs

@@ -0,0 +1,82 @@
+//! Python language syntax implementation.
+
+use crate::{GenericSyntax, LanguageSyntax, escape_python_keyword, to_snake_case};
+
+/// Python-specific code generation syntax.
+pub struct PythonSyntax;
+
+impl LanguageSyntax for PythonSyntax {
+    fn field_name(&self, name: &str) -> String {
+        escape_python_keyword(&to_snake_case(name))
+    }
+
+    fn path_expr(&self, base_var: &str, suffix: &str) -> String {
+        format!("f'{{{}}}{}'", base_var, suffix)
+    }
+
+    fn suffix_expr(&self, acc_var: &str, relative: &str) -> String {
+        if relative.is_empty() {
+            // Identity: just return acc
+            acc_var.to_string()
+        } else {
+            // _m(acc, relative) -> f'{acc}_{relative}' if acc else 'relative'
+            format!("_m({}, '{}')", acc_var, relative)
+        }
+    }
+
+    fn prefix_expr(&self, prefix: &str, acc_var: &str) -> String {
+        if prefix.is_empty() {
+            // Identity: just return acc
+            acc_var.to_string()
+        } else {
+            // _p(prefix, acc) -> f'{prefix}{acc}' if acc else 'prefix_base'
+            let prefix_base = prefix.trim_end_matches('_');
+            format!("_p('{}', {})", prefix_base, acc_var)
+        }
+    }
+
+    fn constructor(&self, type_name: &str, path_expr: &str) -> String {
+        format!("{}(client, {})", type_name, path_expr)
+    }
+
+    fn field_init(&self, indent: &str, name: &str, type_ann: &str, value: &str) -> String {
+        format!("{}self.{}: {} = {}", indent, name, type_ann, value)
+    }
+
+    fn generic_syntax(&self) -> GenericSyntax {
+        GenericSyntax::PYTHON
+    }
+
+    fn string_literal(&self, value: &str) -> String {
+        format!("'{}'", value)
+    }
+
+    fn constructor_name(&self, type_name: &str) -> String {
+        type_name.to_string()
+    }
+
+    fn disc_arg_expr(&self, template: &str) -> String {
+        if template == "{disc}" {
+            "disc".to_string()
+        } else if template.is_empty() {
+            "''".to_string()
+        } else if !template.contains("{disc}") {
+            format!("'{}'", template)
+        } else if template.ends_with("{disc}") {
+            let static_part = template.trim_end_matches("{disc}").trim_end_matches('_');
+            format!("_m('{}', disc)", static_part)
+        } else {
+            format!("f'{}'", template)
+        }
+    }
+
+    fn template_expr(&self, acc_var: &str, template: &str) -> String {
+        if template == "{disc}" {
+            format!("_m({}, disc)", acc_var)
+        } else if !template.contains("{disc}") {
+            format!("_m({}, '{}')", acc_var, template)
+        } else {
+            format!("_m({}, f'{}')", acc_var, template)
+        }
+    }
+}

crates/brk_bindgen/src/backends/rust.rs

@@ -0,0 +1,97 @@
+//! Rust language syntax implementation.
+
+use crate::{GenericSyntax, LanguageSyntax, escape_rust_keyword, to_snake_case};
+
+/// Rust-specific code generation syntax.
+pub struct RustSyntax;
+
+/// Escape braces in a template string for use in `format!()`, preserving `{disc}`.
+fn escape_rust_format(template: &str) -> String {
+    template
+        .replace('{', "{{")
+        .replace('}', "}}")
+        .replace("{{disc}}", "{disc}")
+}
+
+impl LanguageSyntax for RustSyntax {
+    fn field_name(&self, name: &str) -> String {
+        escape_rust_keyword(&to_snake_case(name))
+    }
+
+    fn path_expr(&self, base_var: &str, suffix: &str) -> String {
+        format!("format!(\"{{{}}}{}\")", base_var, suffix)
+    }
+
+    fn suffix_expr(&self, acc_var: &str, relative: &str) -> String {
+        if relative.is_empty() {
+            self.owned_expr(acc_var)
+        } else {
+            format!("_m(&{}, \"{}\")", acc_var, relative)
+        }
+    }
+
+    fn owned_expr(&self, var: &str) -> String {
+        format!("{}.clone()", var)
+    }
+
+    fn prefix_expr(&self, prefix: &str, acc_var: &str) -> String {
+        if prefix.is_empty() {
+            self.owned_expr(acc_var)
+        } else {
+            let prefix_base = prefix.trim_end_matches('_');
+            format!("_p(\"{}\", &{})", prefix_base, acc_var)
+        }
+    }
+
+    fn constructor(&self, type_name: &str, path_expr: &str) -> String {
+        format!("{}::new(client.clone(), {})", type_name, path_expr)
+    }
+
+    fn field_init(&self, indent: &str, name: &str, _type_ann: &str, value: &str) -> String {
+        // Rust struct initialization; type is in struct definition, not in init
+        format!("{}{}: {},", indent, name, value)
+    }
+
+    fn generic_syntax(&self) -> GenericSyntax {
+        GenericSyntax::RUST
+    }
+
+    fn string_literal(&self, value: &str) -> String {
+        format!("\"{}\".to_string()", value)
+    }
+
+    fn constructor_name(&self, type_name: &str) -> String {
+        format!("{}::new", type_name)
+    }
+
+    fn disc_arg_expr(&self, template: &str) -> String {
+        if template == "{disc}" {
+            "disc.clone()".to_string()
+        } else if template.is_empty() {
+            "String::new()".to_string()
+        } else if !template.contains("{disc}") {
+            format!("\"{}\".to_string()", template)
+        } else if template.ends_with("{disc}") {
+            let static_part = template.trim_end_matches("{disc}").trim_end_matches('_');
+            format!("_m(\"{}\", &disc)", static_part)
+        } else {
+            format!("format!(\"{}\")", escape_rust_format(template))
+        }
+    }
+
+    fn template_expr(&self, acc_var: &str, template: &str) -> String {
+        if template == "{disc}" {
+            format!("_m(&{}, &disc)", acc_var)
+        } else if template.is_empty() {
+            acc_var.to_string()
+        } else if !template.contains("{disc}") {
+            format!("_m(&{}, \"{}\")", acc_var, template)
+        } else {
+            format!(
+                "_m(&{}, &format!(\"{}\", disc=disc))",
+                acc_var,
+                escape_rust_format(template)
+            )
+        }
+    }
+}

crates/brk_bindgen/src/generate/constants.rs

@@ -0,0 +1,97 @@
+//! Shared constant generation for static client data.
+//!
+//! Extracts common logic for generating INDEXES, POOL_ID_TO_POOL_NAME,
+//! and cohort name constants across JavaScript and Python clients.
+
+use std::collections::BTreeMap;
+
+use brk_cohort::{
+    AGE_RANGE_NAMES, AMOUNT_RANGE_NAMES, CLASS_NAMES, ENTRY_NAMES, EPOCH_NAMES, LOSS_NAMES,
+    OVER_AGE_NAMES, OVER_AMOUNT_NAMES, PROFIT_NAMES, PROFITABILITY_RANGE_NAMES,
+    SPENDABLE_TYPE_NAMES, TERM_NAMES, UNDER_AGE_NAMES, UNDER_AMOUNT_NAMES,
+};
+use brk_types::{Index, pools};
+use serde::Serialize;
+use serde_json::Value;
+
+use crate::{VERSION, to_camel_case};
+
+/// Collected constant data for client generation.
+pub struct ClientConstants {
+    pub version: String,
+    pub indexes: Vec<&'static str>,
+    pub pool_map: BTreeMap<String, &'static str>,
+}
+
+impl ClientConstants {
+    /// Collect all constant data.
+    pub fn collect() -> Self {
+        let indexes = Index::all();
+        let indexes: Vec<&'static str> = indexes.iter().map(|i| i.name()).collect();
+
+        let pools = pools();
+        let mut sorted_pools: Vec<_> = pools.iter().collect();
+        sorted_pools.sort_by_key(|p| p.name.to_lowercase());
+        let pool_map: BTreeMap<String, &'static str> = sorted_pools
+            .iter()
+            .map(|p| (p.slug().to_string(), p.name))
+            .collect();
+
+        Self {
+            version: format!("v{}", VERSION),
+            indexes,
+            pool_map,
+        }
+    }
+}
+
+/// Cohort name constants - shared data definitions.
+pub struct CohortConstants;
+
+impl CohortConstants {
+    /// Get all cohort constants as name-value pairs for iteration.
+    pub fn all() -> Vec<(&'static str, Value)> {
+        fn to_value<T: Serialize>(v: &T) -> Value {
+            serde_json::to_value(v).unwrap()
+        }
+
+        vec![
+            ("TERM_NAMES", to_value(&TERM_NAMES)),
+            ("EPOCH_NAMES", to_value(&EPOCH_NAMES)),
+            ("CLASS_NAMES", to_value(&CLASS_NAMES)),
+            ("ENTRY_NAMES", to_value(&ENTRY_NAMES)),
+            ("SPENDABLE_TYPE_NAMES", to_value(&SPENDABLE_TYPE_NAMES)),
+            ("AGE_RANGE_NAMES", to_value(&AGE_RANGE_NAMES)),
+            ("UNDER_AGE_NAMES", to_value(&UNDER_AGE_NAMES)),
+            ("OVER_AGE_NAMES", to_value(&OVER_AGE_NAMES)),
+            ("AMOUNT_RANGE_NAMES", to_value(&AMOUNT_RANGE_NAMES)),
+            ("OVER_AMOUNT_NAMES", to_value(&OVER_AMOUNT_NAMES)),
+            ("UNDER_AMOUNT_NAMES", to_value(&UNDER_AMOUNT_NAMES)),
+            (
+                "PROFITABILITY_RANGE_NAMES",
+                to_value(&PROFITABILITY_RANGE_NAMES),
+            ),
+            ("PROFIT_NAMES", to_value(&PROFIT_NAMES)),
+            ("LOSS_NAMES", to_value(&LOSS_NAMES)),
+        ]
+    }
+}
+
+/// Convert top-level keys of a JSON object to camelCase.
+pub fn camel_case_keys(value: Value) -> Value {
+    match value {
+        Value::Object(map) => {
+            let new_map: serde_json::Map<String, Value> = map
+                .into_iter()
+                .map(|(k, v)| (to_camel_case(&k), v))
+                .collect();
+            Value::Object(new_map)
+        }
+        other => other,
+    }
+}
+
+/// Format a JSON value as a pretty-printed string.
+pub fn format_json<T: Serialize>(value: &T) -> String {
+    serde_json::to_string_pretty(value).unwrap()
+}

crates/brk_bindgen/src/generate/fields.rs

@@ -0,0 +1,186 @@
+//! Shared field generation logic.
+//!
+//! This module contains the core field generation logic that is shared
+//! across all language backends. The `LanguageSyntax` trait is used to
+//! abstract over language-specific formatting.
+
+use std::fmt::Write;
+
+use brk_types::SeriesLeafWithSchema;
+
+use crate::{
+    ClientMetadata, LanguageSyntax, PatternBaseResult, PatternField, PatternMode, StructuralPattern,
+};
+
+/// Create a path suffix from a name.
+fn path_suffix(name: &str) -> String {
+    if name.starts_with('_') {
+        name.to_string()
+    } else {
+        format!("_{}", name)
+    }
+}
+
+/// Compute the constructor value for a parameterized field (factory context).
+///
+/// Handles all three pattern modes (Suffix/Prefix/Templated) and the special
+/// case of templated child patterns that need (acc, disc) instead of a path.
+fn compute_parameterized_value<S: LanguageSyntax>(
+    syntax: &S,
+    field: &PatternField,
+    pattern: &StructuralPattern,
+    metadata: &ClientMetadata,
+) -> String {
+    // Templated child patterns receive acc and disc as separate arguments
+    if let Some(child_pattern) = metadata.find_pattern(&field.rust_type)
+        && child_pattern.is_templated()
+    {
+        let disc_template = pattern.get_field_part(&field.name).unwrap_or(&field.name);
+        let disc_arg = syntax.disc_arg_expr(disc_template);
+        let acc_arg = syntax.owned_expr("acc");
+        return syntax.constructor(&field.rust_type, &format!("{acc_arg}, {disc_arg}"));
+    }
+
+    // Compute path expression from pattern mode
+    let path_expr = match pattern.get_field_part(&field.name) {
+        Some(part) => match &pattern.mode {
+            Some(PatternMode::Templated { .. }) => syntax.template_expr("acc", part),
+            Some(PatternMode::Prefix { .. }) => syntax.prefix_expr(part, "acc"),
+            _ => syntax.suffix_expr("acc", part),
+        },
+        None => syntax.path_expr("acc", &path_suffix(&field.name)),
+    };
+
+    // Wrap in constructor — leaves use their index accessor, everything else uses the type name
+    if let Some(accessor) = metadata.find_index_set_pattern(&field.indexes) {
+        syntax.constructor(&accessor.name, &path_expr)
+    } else if field.is_leaf() {
+        panic!(
+            "Field '{}' has no matching index accessor. All series must be indexed.",
+            field.name
+        )
+    } else {
+        syntax.constructor(&field.rust_type, &path_expr)
+    }
+}
+
+/// Generate a parameterized field for a pattern factory.
+///
+/// Used for pattern instances where fields build series names from an accumulated base.
+pub fn generate_parameterized_field<S: LanguageSyntax>(
+    output: &mut String,
+    syntax: &S,
+    field: &PatternField,
+    pattern: &StructuralPattern,
+    metadata: &ClientMetadata,
+    indent: &str,
+) {
+    let field_name = syntax.field_name(&field.name);
+    let type_ann =
+        metadata.field_type_annotation(field, pattern.is_generic, None, syntax.generic_syntax());
+    let value = compute_parameterized_value(syntax, field, pattern, metadata);
+
+    writeln!(
+        output,
+        "{}",
+        syntax.field_init(indent, &field_name, &type_ann, &value)
+    )
+    .unwrap();
+}
+
+/// Generate a tree node field for a pattern-type child.
+///
+/// Called for non-inline branch children that match a parameterizable pattern.
+/// For templated patterns, extracts the discriminator from the base result.
+pub fn generate_tree_node_field<S: LanguageSyntax>(
+    output: &mut String,
+    syntax: &S,
+    field: &PatternField,
+    metadata: &ClientMetadata,
+    indent: &str,
+    client_expr: &str,
+    base_result: &PatternBaseResult,
+) {
+    let field_name = syntax.field_name(&field.name);
+    let type_ann = metadata.field_type_annotation(field, false, None, syntax.generic_syntax());
+    let base_arg = syntax.string_literal(&base_result.base);
+
+    let value = if let Some(pattern) = metadata.find_pattern(&field.rust_type)
+        && pattern.is_templated()
+    {
+        let disc = pattern
+            .extract_disc_from_instance(&base_result.field_parts)
+            .unwrap_or_default();
+        format!(
+            "{}({}, {}, {})",
+            syntax.constructor_name(&field.rust_type),
+            client_expr,
+            base_arg,
+            syntax.string_literal(&disc)
+        )
+    } else {
+        format!(
+            "{}({}, {})",
+            syntax.constructor_name(&field.rust_type),
+            client_expr,
+            base_arg
+        )
+    };
+
+    writeln!(
+        output,
+        "{}",
+        syntax.field_init(indent, &field_name, &type_ann, &value)
+    )
+    .unwrap();
+}
+
+/// Generate a leaf field using the actual series name from the TreeNode::Leaf.
+///
+/// This is the shared implementation for all language backends. It uses
+/// `leaf.name()` directly to get the correct series name, avoiding any
+/// path concatenation that could produce incorrect names.
+///
+/// # Arguments
+/// * `output` - The string buffer to write to
+/// * `syntax` - The language syntax implementation
+/// * `client_expr` - The client expression (e.g., "client.clone()", "this", "client")
+/// * `tree_field_name` - The field name from the tree structure
+/// * `leaf` - The Leaf node containing the actual series name and indexes
+/// * `metadata` - Client metadata for looking up index patterns
+/// * `indent` - Indentation string
+pub fn generate_leaf_field<S: LanguageSyntax>(
+    output: &mut String,
+    syntax: &S,
+    client_expr: &str,
+    tree_field_name: &str,
+    leaf: &SeriesLeafWithSchema,
+    metadata: &ClientMetadata,
+    indent: &str,
+) {
+    let field_name = syntax.field_name(tree_field_name);
+    let accessor = metadata
+        .find_index_set_pattern(leaf.indexes())
+        .unwrap_or_else(|| {
+            panic!(
+                "Series '{}' has no matching index pattern. All series must be indexed.",
+                leaf.name()
+            )
+        });
+
+    let type_ann = metadata.field_type_annotation_from_leaf(leaf, syntax.generic_syntax());
+    let series_name = syntax.string_literal(leaf.name());
+    let value = format!(
+        "{}({}, {})",
+        syntax.constructor_name(&accessor.name),
+        client_expr,
+        series_name
+    );
+
+    writeln!(
+        output,
+        "{}",
+        syntax.field_init(indent, &field_name, &type_ann, &value)
+    )
+    .unwrap();
+}

crates/brk_bindgen/src/generate/mod.rs

@@ -0,0 +1,13 @@
+//! Shared code generation logic.
+//!
+//! This module contains generation functions that are parameterized by
+//! the `LanguageSyntax` trait, allowing them to work across all supported
+//! language backends.
+
+mod constants;
+mod fields;
+mod tree;
+
+pub use constants::*;
+pub use fields::*;
+pub use tree::*;

crates/brk_bindgen/src/generate/tree.rs

@@ -0,0 +1,157 @@
+//! Shared tree generation helpers.
+
+use std::collections::{BTreeMap, BTreeSet};
+
+use brk_types::TreeNode;
+
+use crate::{
+    ClientMetadata, PatternBaseResult, PatternField, child_type_name, get_fields_with_child_info,
+};
+
+/// Build a child path by appending a child name to a parent path.
+/// Uses "/" as separator. If parent is empty, returns just the child name.
+#[inline]
+pub fn build_child_path(parent: &str, child: &str) -> String {
+    if parent.is_empty() {
+        child.to_string()
+    } else {
+        format!("{}/{}", parent, child)
+    }
+}
+
+/// Pre-computed context for a single child node.
+pub struct ChildContext<'a> {
+    /// The child's field name in the tree.
+    pub name: &'a str,
+    /// The child node.
+    pub node: &'a TreeNode,
+    /// The field info for this child (with type_param set for generic patterns).
+    pub field: PatternField,
+    /// Pattern analysis result.
+    pub base_result: PatternBaseResult,
+    /// Whether this is a leaf node.
+    pub is_leaf: bool,
+    /// Whether to use an inline type instead of a pattern type (only meaningful for branches).
+    pub should_inline: bool,
+    /// The type name to use for inline branches.
+    pub inline_type_name: String,
+}
+
+/// Context for generating a tree node, returned by `prepare_tree_node`.
+pub struct TreeNodeContext<'a> {
+    /// Pre-computed context for each child.
+    pub children: Vec<ChildContext<'a>>,
+}
+
+/// Prepare a tree node for generation.
+/// Returns None if the node should be skipped (not a branch, already generated,
+/// or matches a parameterizable pattern).
+///
+/// The `path` parameter is the tree path to this node (e.g., "distribution/utxoCohorts").
+/// It's used to look up pre-computed PatternBaseResult from the analysis phase.
+pub fn prepare_tree_node<'a>(
+    node: &'a TreeNode,
+    name: &str,
+    path: &str,
+    pattern_lookup: &BTreeMap<Vec<PatternField>, String>,
+    metadata: &ClientMetadata,
+    generated: &mut BTreeSet<String>,
+) -> Option<TreeNodeContext<'a>> {
+    let TreeNode::Branch(branch_children) = node else {
+        return None;
+    };
+
+    let fields_with_child_info = get_fields_with_child_info(branch_children, name, pattern_lookup);
+    let fields: Vec<PatternField> = fields_with_child_info
+        .iter()
+        .map(|(f, _)| f.clone())
+        .collect();
+
+    // Look up the pre-computed base result, or use a default that forces inlining
+    let base_result = metadata
+        .get_node_base(path)
+        .cloned()
+        .unwrap_or_else(PatternBaseResult::force_inline);
+
+    // Skip if this matches a parameterizable pattern AND has no outlier AND field parts match
+    let pattern_compatible = pattern_lookup
+        .get(&fields)
+        .and_then(|name| metadata.find_pattern(name))
+        .is_none_or(|p| {
+            p.is_suffix_mode() == base_result.is_suffix_mode
+                && p.field_parts_match(&base_result.field_parts)
+        });
+    if let Some(pattern_name) = pattern_lookup.get(&fields)
+        && pattern_name != name
+        && metadata.is_parameterizable(pattern_name)
+        && !base_result.has_outlier
+        && pattern_compatible
+    {
+        return None;
+    }
+
+    // Skip if already generated
+    if generated.contains(name) {
+        return None;
+    }
+    generated.insert(name.to_string());
+
+    // Build child contexts with pre-computed decisions
+    let children: Vec<ChildContext<'a>> = branch_children
+        .iter()
+        .zip(fields_with_child_info)
+        .map(|((child_name, child_node), (mut field, child_fields))| {
+            let is_leaf = matches!(child_node, TreeNode::Leaf(_));
+
+            // Set type_param for generic patterns so field_type_annotation works directly
+            if let Some(cf) = &child_fields {
+                field.type_param = metadata.get_type_param(cf).cloned();
+            }
+
+            // Build child path and look up its pre-computed base result
+            let child_path = build_child_path(path, child_name);
+            let base_result = metadata
+                .get_node_base(&child_path)
+                .cloned()
+                .unwrap_or_else(PatternBaseResult::force_inline);
+
+            // Single lookup for the child's matching pattern (avoids repeated scans)
+            let matching_pattern = child_fields
+                .as_ref()
+                .and_then(|cf| metadata.find_pattern_by_fields(cf));
+
+            let matches_any_pattern = matching_pattern.is_some();
+            let pattern_compatible = matching_pattern.is_none_or(|p| {
+                p.is_suffix_mode() == base_result.is_suffix_mode
+                    && p.field_parts_match(&base_result.field_parts)
+            });
+            let is_parameterizable =
+                matching_pattern.is_none_or(|p| metadata.is_parameterizable(&p.name));
+
+            // should_inline determines if we generate an inline struct type
+            let should_inline = !is_leaf
+                && (!matches_any_pattern
+                    || !pattern_compatible
+                    || !is_parameterizable
+                    || base_result.has_outlier);
+
+            let inline_type_name = if should_inline {
+                child_type_name(name, child_name)
+            } else {
+                String::new()
+            };
+
+            ChildContext {
+                name: child_name,
+                node: child_node,
+                field,
+                base_result,
+                is_leaf,
+                should_inline,
+                inline_type_name,
+            }
+        })
+        .collect();
+
+    Some(TreeNodeContext { children })
+}

crates/brk_bindgen/src/generators/javascript/api.rs

@@ -0,0 +1,300 @@
+//! JavaScript API method generation.
+
+use std::fmt::Write;
+
+use crate::{
+    Endpoint, Parameter,
+    generators::{javascript::types::jsdoc_normalize, normalize_return_type, write_description},
+    to_camel_case,
+};
+
+/// Generate API methods for the BrkClient class.
+pub fn generate_api_methods(output: &mut String, endpoints: &[Endpoint]) {
+    for endpoint in endpoints {
+        if !endpoint.should_generate() {
+            continue;
+        }
+        match endpoint.method.as_str() {
+            "GET" => generate_get_method(output, endpoint),
+            "POST" => generate_post_method(output, endpoint),
+            _ => continue,
+        }
+    }
+}
+
+fn generate_get_method(output: &mut String, endpoint: &Endpoint) {
+    let method_name = endpoint_to_method_name(endpoint);
+    let return_type = build_return_type(endpoint);
+
+    write_method_doc(output, endpoint);
+    for param in &endpoint.path_params {
+        let desc = format_param_desc(param.description.as_deref());
+        let ty = jsdoc_normalize(&param.param_type);
+        writeln!(output, "   * @param {{{}}} {}{}", ty, param.name, desc).unwrap();
+    }
+    for param in &endpoint.query_params {
+        let optional = if param.required { "" } else { "=" };
+        let desc = format_param_desc(param.description.as_deref());
+        let ty = jsdoc_normalize(&param.param_type);
+        let ident = sanitize_ident(&param.name);
+        let name_decl = if param.required {
+            ident
+        } else {
+            format!("[{}]", ident)
+        };
+        writeln!(
+            output,
+            "   * @param {{{}{}}} {}{}",
+            ty, optional, name_decl, desc
+        )
+        .unwrap();
+    }
+    writeln!(
+        output,
+        "   * @param {{{{ signal?: AbortSignal, onValue?: (value: {}) => void, cache?: boolean }}}} [options]",
+        return_type
+    )
+    .unwrap();
+    writeln!(output, "   * @returns {{Promise<{}>}}", return_type).unwrap();
+    writeln!(output, "   */").unwrap();
+
+    let params = build_method_params(endpoint);
+    let params_with_opts = if params.is_empty() {
+        "{ signal, onValue, cache } = {}".to_string()
+    } else {
+        format!("{}, {{ signal, onValue, cache }} = {{}}", params)
+    };
+    writeln!(output, "  async {}({}) {{", method_name, params_with_opts).unwrap();
+
+    let path = build_path_template(&endpoint.path, &endpoint.path_params);
+
+    let fetch_call: String = if endpoint.returns_binary() {
+        "this.getBytes(path, { signal, onValue, cache })".to_string()
+    } else if endpoint.returns_json() {
+        "this.getJson(path, { signal, onValue, cache })".to_string()
+    } else if endpoint.response_kind.text_is_numeric() {
+        "Number(await this.getText(path, { signal, cache, onValue: onValue ? (v) => onValue(Number(v)) : undefined }))".to_string()
+    } else {
+        "this.getText(path, { signal, onValue, cache })".to_string()
+    };
+
+    write_path_assignment(output, endpoint, &path);
+
+    if endpoint.supports_csv {
+        writeln!(
+            output,
+            "    if (format === 'csv') return this.getText(path, {{ signal, onValue, cache }});"
+        )
+        .unwrap();
+    }
+    writeln!(output, "    return {};", fetch_call).unwrap();
+    writeln!(output, "  }}\n").unwrap();
+}
+
+fn generate_post_method(output: &mut String, endpoint: &Endpoint) {
+    let method_name = endpoint_to_method_name(endpoint);
+    let return_type = build_return_type(endpoint);
+
+    write_method_doc(output, endpoint);
+    for param in &endpoint.path_params {
+        let desc = format_param_desc(param.description.as_deref());
+        let ty = jsdoc_normalize(&param.param_type);
+        writeln!(output, "   * @param {{{}}} {}{}", ty, param.name, desc).unwrap();
+    }
+    for param in &endpoint.query_params {
+        let optional = if param.required { "" } else { "=" };
+        let desc = format_param_desc(param.description.as_deref());
+        let ty = jsdoc_normalize(&param.param_type);
+        writeln!(
+            output,
+            "   * @param {{{}{}}} [{}]{}",
+            ty, optional, param.name, desc
+        )
+        .unwrap();
+    }
+    if let Some(body) = &endpoint.request_body {
+        let optional = if body.required { "" } else { "=" };
+        let ty = jsdoc_normalize(&body.body_type);
+        writeln!(
+            output,
+            "   * @param {{{}{}}} body - Request body",
+            ty, optional
+        )
+        .unwrap();
+    }
+    writeln!(
+        output,
+        "   * @param {{{{ signal?: AbortSignal }}}} [options]"
+    )
+    .unwrap();
+    writeln!(output, "   * @returns {{Promise<{}>}}", return_type).unwrap();
+    writeln!(output, "   */").unwrap();
+
+    let mut params = build_method_params(endpoint);
+    if endpoint.request_body.is_some() {
+        if !params.is_empty() {
+            params.push_str(", ");
+        }
+        params.push_str("body");
+    }
+    let params_with_opts = if params.is_empty() {
+        "{ signal } = {}".to_string()
+    } else {
+        format!("{}, {{ signal }} = {{}}", params)
+    };
+    writeln!(output, "  async {}({}) {{", method_name, params_with_opts).unwrap();
+
+    let path = build_path_template(&endpoint.path, &endpoint.path_params);
+    let body_arg = if endpoint.request_body.is_some() {
+        "body"
+    } else {
+        "''"
+    };
+
+    let fetch_call: String = if endpoint.returns_binary() {
+        format!("this.postBytes(path, {}, {{ signal }})", body_arg)
+    } else if endpoint.returns_json() {
+        format!("this.postJson(path, {}, {{ signal }})", body_arg)
+    } else if endpoint.response_kind.text_is_numeric() {
+        format!(
+            "Number(await this.postText(path, {}, {{ signal }}))",
+            body_arg
+        )
+    } else {
+        format!("this.postText(path, {}, {{ signal }})", body_arg)
+    };
+
+    write_path_assignment(output, endpoint, &path);
+
+    writeln!(output, "    return {};", fetch_call).unwrap();
+    writeln!(output, "  }}\n").unwrap();
+}
+
+fn build_return_type(endpoint: &Endpoint) -> String {
+    let base = if endpoint.returns_binary() {
+        "Uint8Array".to_string()
+    } else {
+        jsdoc_normalize(&normalize_return_type(
+            endpoint.schema_name().unwrap_or("*"),
+        ))
+    };
+    if endpoint.supports_csv {
+        format!("{} | string", base)
+    } else {
+        base
+    }
+}
+
+fn write_method_doc(output: &mut String, endpoint: &Endpoint) {
+    writeln!(output, "  /**").unwrap();
+    if let Some(summary) = &endpoint.summary {
+        writeln!(output, "   * {}", summary).unwrap();
+    }
+    if let Some(desc) = &endpoint.description
+        && endpoint.summary.as_ref() != Some(desc)
+    {
+        writeln!(output, "   *").unwrap();
+        write_description(output, desc, "   * ", "   *");
+    }
+    writeln!(output, "   *").unwrap();
+    writeln!(
+        output,
+        "   * Endpoint: `{} {}`",
+        endpoint.method.to_uppercase(),
+        endpoint.path
+    )
+    .unwrap();
+
+    let has_body_param = endpoint.method == "POST" && endpoint.request_body.is_some();
+    if !endpoint.path_params.is_empty() || !endpoint.query_params.is_empty() || has_body_param {
+        writeln!(output, "   *").unwrap();
+    }
+}
+
+fn write_path_assignment(output: &mut String, endpoint: &Endpoint, path: &str) {
+    if endpoint.query_params.is_empty() {
+        writeln!(output, "    const path = `{}`;", path).unwrap();
+    } else {
+        writeln!(output, "    const params = new URLSearchParams();").unwrap();
+        for param in &endpoint.query_params {
+            let ident = sanitize_ident(&param.name);
+            let is_array = param.param_type.ends_with("[]");
+            if is_array {
+                if param.required {
+                    writeln!(
+                        output,
+                        "    for (const _v of {}) params.append('{}', String(_v));",
+                        ident, param.name
+                    )
+                    .unwrap();
+                } else {
+                    writeln!(
+                        output,
+                        "    if ({}) for (const _v of {}) params.append('{}', String(_v));",
+                        ident, ident, param.name
+                    )
+                    .unwrap();
+                }
+            } else if param.required {
+                writeln!(
+                    output,
+                    "    params.set('{}', String({}));",
+                    param.name, ident
+                )
+                .unwrap();
+            } else {
+                writeln!(
+                    output,
+                    "    if ({} !== undefined) params.set('{}', String({}));",
+                    ident, param.name, ident
+                )
+                .unwrap();
+            }
+        }
+        writeln!(output, "    const query = params.toString();").unwrap();
+        writeln!(
+            output,
+            "    const path = `{}${{query ? '?' + query : ''}}`;",
+            path
+        )
+        .unwrap();
+    }
+}
+
+fn endpoint_to_method_name(endpoint: &Endpoint) -> String {
+    to_camel_case(&endpoint.operation_name())
+}
+
+fn build_method_params(endpoint: &Endpoint) -> String {
+    let mut params = Vec::new();
+    for param in &endpoint.path_params {
+        params.push(sanitize_ident(&param.name));
+    }
+    for param in &endpoint.query_params {
+        params.push(sanitize_ident(&param.name));
+    }
+    params.join(", ")
+}
+
+/// Strip characters invalid in JS identifiers (e.g. `[]` from `txId[]`).
+fn sanitize_ident(name: &str) -> String {
+    name.replace(['[', ']'], "")
+}
+
+fn build_path_template(path: &str, path_params: &[Parameter]) -> String {
+    let mut result = path.to_string();
+    for param in path_params {
+        let placeholder = format!("{{{}}}", param.name);
+        let interpolation = format!("${{{}}}", param.name);
+        result = result.replace(&placeholder, &interpolation);
+    }
+    result
+}
+
+/// Format param description with dash prefix, or empty string if no description.
+fn format_param_desc(desc: Option<&str>) -> String {
+    match desc {
+        Some(d) if !d.is_empty() => format!(" - {}", d),
+        _ => String::new(),
+    }
+}

crates/brk_bindgen/src/generators/javascript/client.rs

@@ -0,0 +1,938 @@
+//! JavaScript base client and pattern factory generation.
+
+use std::fmt::Write;
+
+use crate::{
+    ClientConstants, ClientMetadata, CohortConstants, GenericSyntax, IndexSetPattern,
+    JavaScriptSyntax, StructuralPattern, camel_case_keys, format_json,
+    generate_parameterized_field, to_camel_case,
+};
+
+/// Generate the base BrkClient class with HTTP functionality.
+pub fn generate_base_client(output: &mut String) {
+    writeln!(
+        output,
+        r#"/**
+ * @typedef {{Object}} BrkClientOptions
+ * @property {{string}} baseUrl - Base URL for the API
+ * @property {{number}} [timeout] - Request timeout in milliseconds
+ * @property {{string|boolean}} [browserCache] - Enable browser Cache API with default name (true), custom name (string), or disable (false). No effect in Node.js. Default: true
+ * @property {{number|boolean}} [memCache] - In-memory parsed-response cache size (LRU). true/undefined → 1000, false/0 → disabled. Lets 304 responses skip the JSON parse entirely. Default: 1000
+ */
+
+const _isBrowser = typeof window !== 'undefined' && 'caches' in window;
+const _runIdle = (/** @type {{VoidFunction}} */ fn) => (globalThis.requestIdleCallback ?? setTimeout)(fn);
+const _defaultBrowserCacheName = '__BRK_CLIENT__';
+const _DEFAULT_MEM_CACHE_SIZE = 1000;
+
+/** @template T @typedef {{{{ etag: string | null, value: T }}}} _MemEntry */
+/** @param {{*}} v */
+const _addCamelGetters = (v) => {{
+  if (Array.isArray(v)) {{ v.forEach(_addCamelGetters); return v; }}
+  if (v && typeof v === 'object' && v.constructor === Object) {{
+    for (const k in v) {{
+      if (k.includes('_')) {{
+        const c = k.replace(/_([a-z])/g, (_, l) => l.toUpperCase());
+        if (!(c in v)) Object.defineProperty(v, c, {{ get() {{ return this[k]; }} }});
+      }}
+      _addCamelGetters(v[k]);
+    }}
+  }}
+  return v;
+}};
+
+/**
+ * @param {{string|boolean|undefined}} option
+ * @returns {{Promise<Cache | null>}}
+ */
+const _openBrowserCache = (option) => {{
+  if (!_isBrowser || option === false) return Promise.resolve(null);
+  const name = typeof option === 'string' ? option : _defaultBrowserCacheName;
+  return caches.open(name).catch(() => null);
+}};
+
+/**
+ * @param {{string}} url
+ * @returns {{URL}}
+ */
+const _parseBaseUrl = (url) => new URL(url, typeof location === 'undefined' ? undefined : location.href);
+
+/**
+ * Custom error class for BRK client errors
+ */
+class BrkError extends Error {{
+  /**
+   * @param {{string}} message
+   * @param {{number}} [status]
+   */
+  constructor(message, status) {{
+    super(message);
+    this.name = 'BrkError';
+    this.status = status;
+  }}
+}}
+
+// Date conversion constants and helpers
+const _GENESIS = new Date(2009, 0, 3);  // day1 0, week1 0
+const _DAY_ONE = new Date(2009, 0, 9);  // day1 1 (6 day gap after genesis)
+const _MS_PER_DAY = 86400000;
+const _MS_PER_WEEK = 7 * _MS_PER_DAY;
+const _EPOCH_MS = 1230768000000;
+const _DATE_INDEXES = new Set([
+  'minute10', 'minute30',
+  'hour1', 'hour4', 'hour12',
+  'day1', 'day3', 'week1',
+  'month1', 'month3', 'month6',
+  'year1', 'year10',
+]);
+
+/** @param {{number}} months @returns {{globalThis.Date}} */
+const _addMonths = (months) => new Date(2009, months, 1);
+
+/**
+ * Convert an index value to a Date for date-based indexes.
+ * @param {{Index}} index - The index type
+ * @param {{number}} i - The index value
+ * @returns {{globalThis.Date}}
+ */
+function indexToDate(index, i) {{
+  switch (index) {{
+    case 'minute10': return new Date(_EPOCH_MS + i * 600000);
+    case 'minute30': return new Date(_EPOCH_MS + i * 1800000);
+    case 'hour1': return new Date(_EPOCH_MS + i * 3600000);
+    case 'hour4': return new Date(_EPOCH_MS + i * 14400000);
+    case 'hour12': return new Date(_EPOCH_MS + i * 43200000);
+    case 'day1': return i === 0 ? _GENESIS : new Date(_DAY_ONE.getTime() + (i - 1) * _MS_PER_DAY);
+    case 'day3': return new Date(_EPOCH_MS - 86400000 + i * 259200000);
+    case 'week1': return new Date(_GENESIS.getTime() + i * _MS_PER_WEEK);
+    case 'month1': return _addMonths(i);
+    case 'month3': return _addMonths(i * 3);
+    case 'month6': return _addMonths(i * 6);
+    case 'year1': return new Date(2009 + i, 0, 1);
+    case 'year10': return new Date(2009 + i * 10, 0, 1);
+    default: throw new Error(`${{index}} is not a date-based index`);
+  }}
+}}
+
+/**
+ * Convert a Date to an index value for date-based indexes.
+ * Returns the floor index (latest index whose date is <= the given date).
+ * @param {{Index}} index - The index type
+ * @param {{globalThis.Date}} d - The date to convert
+ * @returns {{number}}
+ */
+function dateToIndex(index, d) {{
+  const ms = d.getTime();
+  switch (index) {{
+    case 'minute10': return Math.floor((ms - _EPOCH_MS) / 600000);
+    case 'minute30': return Math.floor((ms - _EPOCH_MS) / 1800000);
+    case 'hour1': return Math.floor((ms - _EPOCH_MS) / 3600000);
+    case 'hour4': return Math.floor((ms - _EPOCH_MS) / 14400000);
+    case 'hour12': return Math.floor((ms - _EPOCH_MS) / 43200000);
+    case 'day1': {{
+      if (ms < _DAY_ONE.getTime()) return 0;
+      return 1 + Math.floor((ms - _DAY_ONE.getTime()) / _MS_PER_DAY);
+    }}
+    case 'day3': return Math.floor((ms - _EPOCH_MS + 86400000) / 259200000);
+    case 'week1': return Math.floor((ms - _GENESIS.getTime()) / _MS_PER_WEEK);
+    case 'month1': return (d.getFullYear() - 2009) * 12 + d.getMonth();
+    case 'month3': return (d.getFullYear() - 2009) * 4 + Math.floor(d.getMonth() / 3);
+    case 'month6': return (d.getFullYear() - 2009) * 2 + Math.floor(d.getMonth() / 6);
+    case 'year1': return d.getFullYear() - 2009;
+    case 'year10': return Math.floor((d.getFullYear() - 2009) / 10);
+    default: throw new Error(`${{index}} is not a date-based index`);
+  }}
+}}
+
+/**
+ * Wrap raw series data with helper methods.
+ * @template T
+ * @param {{SeriesData<T>}} raw - Raw JSON response
+ * @returns {{DateSeriesData<T>}}
+ */
+function _wrapSeriesData(raw) {{
+  const {{ index, start, end, data }} = raw;
+  const _dateBased = _DATE_INDEXES.has(index);
+  return /** @type {{DateSeriesData<T>}} */ ({{
+    ...raw,
+    isDateBased: _dateBased,
+    indexes() {{
+      /** @type {{number[]}} */
+      const result = [];
+      for (let i = start; i < end; i++) result.push(i);
+      return result;
+    }},
+    keys() {{
+      return this.indexes();
+    }},
+    entries() {{
+      /** @type {{Array<[number, T]>}} */
+      const result = [];
+      for (let i = 0; i < data.length; i++) result.push([start + i, data[i]]);
+      return result;
+    }},
+    toMap() {{
+      /** @type {{Map<number, T>}} */
+      const map = new Map();
+      for (let i = 0; i < data.length; i++) map.set(start + i, data[i]);
+      return map;
+    }},
+    *[Symbol.iterator]() {{
+      for (let i = 0; i < data.length; i++) yield /** @type {{[number, T]}} */ ([start + i, data[i]]);
+    }},
+    // DateSeriesData methods (only meaningful for date-based indexes)
+    dates() {{
+      /** @type {{globalThis.Date[]}} */
+      const result = [];
+      for (let i = start; i < end; i++) result.push(indexToDate(index, i));
+      return result;
+    }},
+    dateEntries() {{
+      /** @type {{Array<[globalThis.Date, T]>}} */
+      const result = [];
+      for (let i = 0; i < data.length; i++) result.push([indexToDate(index, start + i), data[i]]);
+      return result;
+    }},
+    toDateMap() {{
+      /** @type {{Map<globalThis.Date, T>}} */
+      const map = new Map();
+      for (let i = 0; i < data.length; i++) map.set(indexToDate(index, start + i), data[i]);
+      return map;
+    }},
+  }});
+}}
+
+/**
+ * @template T
+ * @typedef {{Object}} SeriesDataBase
+ * @property {{number}} version - Version of the series data
+ * @property {{Index}} index - The index type used for this query
+ * @property {{string}} type - Value type (e.g. "f32", "u64", "Sats")
+ * @property {{number}} start - Start index (inclusive)
+ * @property {{number}} end - End index (exclusive)
+ * @property {{string}} stamp - ISO 8601 timestamp of when the response was generated
+ * @property {{T[]}} data - The series data
+ * @property {{boolean}} isDateBased - Whether this series uses a date-based index
+ * @property {{() => number[]}} indexes - Get index numbers
+ * @property {{() => number[]}} keys - Get keys as index numbers (alias for indexes)
+ * @property {{() => Array<[number, T]>}} entries - Get [index, value] pairs
+ * @property {{() => Map<number, T>}} toMap - Convert to Map<index, value>
+ */
+
+/** @template T @typedef {{SeriesDataBase<T> & Iterable<[number, T]>}} SeriesData */
+
+/**
+ * @template T
+ * @typedef {{Object}} DateSeriesDataExtras
+ * @property {{() => globalThis.Date[]}} dates - Get dates for each data point
+ * @property {{() => Array<[globalThis.Date, T]>}} dateEntries - Get [date, value] pairs
+ * @property {{() => Map<globalThis.Date, T>}} toDateMap - Convert to Map<date, value>
+ */
+
+/** @template T @typedef {{SeriesData<T> & DateSeriesDataExtras<T>}} DateSeriesData */
+/** @typedef {{SeriesData<any>}} AnySeriesData */
+
+/** @template T @typedef {{(onfulfilled?: (value: SeriesData<T>) => any, onrejected?: (reason: Error) => never) => Promise<SeriesData<T>>}} Thenable */
+/** @template T @typedef {{(onfulfilled?: (value: DateSeriesData<T>) => any, onrejected?: (reason: Error) => never) => Promise<DateSeriesData<T>>}} DateThenable */
+
+/**
+ * @template T
+ * @typedef {{Object}} SeriesEndpoint
+ * @property {{(index: number) => SingleItemBuilder<T>}} get - Get single item at index
+ * @property {{(start?: number, end?: number) => RangeBuilder<T>}} slice - Slice by index
+ * @property {{(n: number) => RangeBuilder<T>}} first - Get first n items
+ * @property {{(n: number) => RangeBuilder<T>}} last - Get last n items
+ * @property {{(n: number) => SkippedBuilder<T>}} skip - Skip first n items, chain with take()
+ * @property {{(onValue?: (value: SeriesData<T>) => void) => Promise<SeriesData<T>>}} fetch - Fetch all data
+ * @property {{() => Promise<string>}} fetchCsv - Fetch all data as CSV
+ * @property {{() => Promise<number>}} len - Get total number of data points
+ * @property {{() => Promise<Version>}} version - Get the current version of the series
+ * @property {{Thenable<T>}} then - Thenable (await endpoint)
+ * @property {{string}} path - The endpoint path
+ */
+
+/**
+ * @template T
+ * @typedef {{Object}} DateSeriesEndpoint
+ * @property {{(index: number | globalThis.Date) => DateSingleItemBuilder<T>}} get - Get single item at index or Date
+ * @property {{(start?: number | globalThis.Date, end?: number | globalThis.Date) => DateRangeBuilder<T>}} slice - Slice by index or Date
+ * @property {{(n: number) => DateRangeBuilder<T>}} first - Get first n items
+ * @property {{(n: number) => DateRangeBuilder<T>}} last - Get last n items
+ * @property {{(n: number) => DateSkippedBuilder<T>}} skip - Skip first n items, chain with take()
+ * @property {{(onValue?: (value: DateSeriesData<T>) => void) => Promise<DateSeriesData<T>>}} fetch - Fetch all data
+ * @property {{() => Promise<string>}} fetchCsv - Fetch all data as CSV
+ * @property {{() => Promise<number>}} len - Get total number of data points
+ * @property {{() => Promise<Version>}} version - Get the current version of the series
+ * @property {{DateThenable<T>}} then - Thenable (await endpoint)
+ * @property {{string}} path - The endpoint path
+ */
+
+/** @typedef {{SeriesEndpoint<any>}} AnySeriesEndpoint */
+
+/** @template T @typedef {{Object}} SingleItemBuilder
+ * @property {{(onValue?: (value: SeriesData<T>) => void) => Promise<SeriesData<T>>}} fetch - Fetch the item
+ * @property {{() => Promise<string>}} fetchCsv - Fetch as CSV
+ * @property {{Thenable<T>}} then - Thenable
+ */
+
+/** @template T @typedef {{Object}} DateSingleItemBuilder
+ * @property {{(onValue?: (value: DateSeriesData<T>) => void) => Promise<DateSeriesData<T>>}} fetch - Fetch the item
+ * @property {{() => Promise<string>}} fetchCsv - Fetch as CSV
+ * @property {{DateThenable<T>}} then - Thenable
+ */
+
+/** @template T @typedef {{Object}} SkippedBuilder
+ * @property {{(n: number) => RangeBuilder<T>}} take - Take n items after skipped position
+ * @property {{(onValue?: (value: SeriesData<T>) => void) => Promise<SeriesData<T>>}} fetch - Fetch from skipped position to end
+ * @property {{() => Promise<string>}} fetchCsv - Fetch as CSV
+ * @property {{Thenable<T>}} then - Thenable
+ */
+
+/** @template T @typedef {{Object}} DateSkippedBuilder
+ * @property {{(n: number) => DateRangeBuilder<T>}} take - Take n items after skipped position
+ * @property {{(onValue?: (value: DateSeriesData<T>) => void) => Promise<DateSeriesData<T>>}} fetch - Fetch from skipped position to end
+ * @property {{() => Promise<string>}} fetchCsv - Fetch as CSV
+ * @property {{DateThenable<T>}} then - Thenable
+ */
+
+/** @template T @typedef {{Object}} RangeBuilder
+ * @property {{(onValue?: (value: SeriesData<T>) => void) => Promise<SeriesData<T>>}} fetch - Fetch the range
+ * @property {{() => Promise<string>}} fetchCsv - Fetch as CSV
+ * @property {{Thenable<T>}} then - Thenable
+ */
+
+/** @template T @typedef {{Object}} DateRangeBuilder
+ * @property {{(onValue?: (value: DateSeriesData<T>) => void) => Promise<DateSeriesData<T>>}} fetch - Fetch the range
+ * @property {{() => Promise<string>}} fetchCsv - Fetch as CSV
+ * @property {{DateThenable<T>}} then - Thenable
+ */
+
+/**
+ * @template T
+ * @typedef {{Object}} SeriesPattern
+ * @property {{string}} name - The series name
+ * @property {{Readonly<Partial<Record<Index, SeriesEndpoint<T>>>>}} by - Index endpoints as lazy getters
+ * @property {{() => readonly Index[]}} indexes - Get the list of available indexes
+ * @property {{(index: Index) => SeriesEndpoint<T>|undefined}} get - Get an endpoint for a specific index
+ */
+
+/** @typedef {{SeriesPattern<any>}} AnySeriesPattern */
+
+/**
+ * Create a series endpoint builder with typestate pattern.
+ * @template T
+ * @param {{BrkClient}} client
+ * @param {{string}} name - The series vec name
+ * @param {{Index}} index - The index name
+ * @returns {{DateSeriesEndpoint<T>}}
+ */
+function _endpoint(client, name, index) {{
+  const p = `/api/series/${{name}}/${{index}}`;
+
+  /**
+   * @param {{number}} [start]
+   * @param {{number}} [end]
+   * @param {{string}} [format]
+   * @returns {{string}}
+   */
+  const buildPath = (start, end, format) => {{
+    const params = new URLSearchParams();
+    if (start !== undefined) params.set('start', String(start));
+    if (end !== undefined) params.set('end', String(end));
+    if (format) params.set('format', format);
+    const query = params.toString();
+    return query ? `${{p}}?${{query}}` : p;
+  }};
+
+  /**
+   * @param {{number}} [start]
+   * @param {{number}} [end]
+   * @returns {{DateRangeBuilder<T>}}
+   */
+  const rangeBuilder = (start, end) => ({{
+    fetch(onValue) {{ return client._fetchSeriesData(buildPath(start, end), onValue); }},
+    fetchCsv() {{ return client.getText(buildPath(start, end, 'csv')); }},
+    then(resolve, reject) {{ return this.fetch().then(resolve, reject); }},
+  }});
+
+  /**
+   * @param {{number}} idx
+   * @returns {{DateSingleItemBuilder<T>}}
+   */
+  const singleItemBuilder = (idx) => ({{
+    fetch(onValue) {{ return client._fetchSeriesData(buildPath(idx, idx + 1), onValue); }},
+    fetchCsv() {{ return client.getText(buildPath(idx, idx + 1, 'csv')); }},
+    then(resolve, reject) {{ return this.fetch().then(resolve, reject); }},
+  }});
+
+  /**
+   * @param {{number}} start
+   * @returns {{DateSkippedBuilder<T>}}
+   */
+  const skippedBuilder = (start) => ({{
+    take(n) {{ return rangeBuilder(start, start + n); }},
+    fetch(onValue) {{ return client._fetchSeriesData(buildPath(start, undefined), onValue); }},
+    fetchCsv() {{ return client.getText(buildPath(start, undefined, 'csv')); }},
+    then(resolve, reject) {{ return this.fetch().then(resolve, reject); }},
+  }});
+
+  /** @type {{DateSeriesEndpoint<T>}} */
+  const endpoint = {{
+    get(idx) {{ if (idx instanceof Date) idx = dateToIndex(index, idx); return singleItemBuilder(idx); }},
+    slice(start, end) {{
+      if (start instanceof Date) start = dateToIndex(index, start);
+      if (end instanceof Date) end = dateToIndex(index, end);
+      return rangeBuilder(start, end);
+    }},
+    first(n) {{ return rangeBuilder(undefined, n); }},
+    last(n) {{ return n === 0 ? rangeBuilder(undefined, 0) : rangeBuilder(-n, undefined); }},
+    skip(n) {{ return skippedBuilder(n); }},
+    fetch(onValue) {{ return client._fetchSeriesData(buildPath(), onValue); }},
+    fetchCsv() {{ return client.getText(buildPath(undefined, undefined, 'csv')); }},
+    len() {{ return client.getSeriesLen(name, index); }},
+    version() {{ return client.getSeriesVersion(name, index); }},
+    then(resolve, reject) {{ return this.fetch().then(resolve, reject); }},
+    get path() {{ return p; }},
+  }};
+
+  return endpoint;
+}}
+
+/**
+ * Base HTTP client for making requests with caching support
+ */
+class BrkClientBase {{
+  /**
+   * @param {{BrkClientOptions|string}} options
+   */
+  constructor(options) {{
+    const isString = typeof options === 'string';
+    const rawUrl = isString ? options : options.baseUrl;
+    this.baseUrl = rawUrl.endsWith('/') ? rawUrl.slice(0, -1) : rawUrl;
+    const url = _parseBaseUrl(this.baseUrl);
+    this.url = url.href.endsWith('/') ? url.href.slice(0, -1) : url.href;
+    this.domain = url.hostname;
+    this.timeout = isString ? 5000 : (options.timeout ?? 5000);
+    /** @type {{Promise<Cache | null>}} */
+    this._browserCachePromise = _openBrowserCache(isString ? undefined : options.browserCache);
+    /** @type {{Cache | null}} */
+    this._browserCache = null;
+    this._browserCachePromise.then(c => this._browserCache = c);
+    const memOpt = isString ? undefined : options.memCache;
+    this._memCacheMax = memOpt === false || memOpt === 0
+      ? 0
+      : (typeof memOpt === 'number' ? memOpt : _DEFAULT_MEM_CACHE_SIZE);
+    /** @type {{Map<string, _MemEntry<unknown>>}} */
+    this._memCache = new Map();
+  }}
+
+  /**
+   * @template T
+   * @param {{string}} key
+   * @returns {{_MemEntry<T> | undefined}}
+   */
+  _memGet(key) {{
+    if (!this._memCacheMax) return undefined;
+    const hit = this._memCache.get(key);
+    if (!hit) return undefined;
+    this._memCache.delete(key);
+    this._memCache.set(key, hit);
+    return /** @type {{_MemEntry<T>}} */ (hit);
+  }}
+
+  /**
+   * @param {{string}} key
+   * @param {{string | null}} etag
+   * @param {{unknown}} value
+   */
+  _memSet(key, etag, value) {{
+    if (!this._memCacheMax) return;
+    if (this._memCache.has(key)) this._memCache.delete(key);
+    else if (this._memCache.size >= this._memCacheMax) {{
+      const oldest = this._memCache.keys().next().value;
+      if (oldest !== undefined) this._memCache.delete(oldest);
+    }}
+    this._memCache.set(key, {{ etag, value }});
+  }}
+
+  /**
+   * @param {{string}} path
+   * @param {{{{ signal?: AbortSignal, cache?: boolean }}}} [options]
+   * @returns {{Promise<Response>}}
+   */
+  async get(path, {{ signal, cache = true }} = {{}}) {{
+    const url = `${{this.baseUrl}}${{path}}`;
+    const signals = [AbortSignal.timeout(this.timeout)];
+    if (signal) signals.push(signal);
+    /** @type {{RequestInit}} */
+    const init = {{ signal: AbortSignal.any(signals) }};
+    if (!cache) init.cache = 'no-store';
+    const res = await fetch(url, init);
+    if (!res.ok) throw new BrkError(`HTTP ${{res.status}}: ${{url}}`, res.status);
+    return res;
+  }}
+
+  /**
+   * Make a GET request with layered caching.
+   *
+   * Contract:
+   * - The returned Promise resolves with the **freshest** value (post-revalidation).
+   * - `onValue` fires once with the freshest value, or twice if a stale snapshot
+   *   could be shown first (stale-while-revalidate). On a 304 there is no second fire.
+   *
+   * Layers:
+   * - L1 (memCache): in-memory parsed values keyed by URL+ETag. Lets 304s skip the parse entirely.
+   * - L2 (browserCache): Cache API, survives reload and feeds onValue fast on cold start.
+   *
+   * @template T
+   * @param {{string}} path
+   * @param {{(res: Response) => Promise<T>}} parse - Response body reader
+   * @param {{{{ onValue?: (value: T) => void, signal?: AbortSignal, cache?: boolean }}}} [options]
+   * @returns {{Promise<T>}}
+   */
+  async _getCached(path, parse, {{ onValue, signal, cache = true }} = {{}}) {{
+    if (!cache) {{
+      const res = await this.get(path, {{ signal, cache }});
+      const value = await parse(res);
+      if (onValue) onValue(value);
+      return value;
+    }}
+
+    const url = `${{this.baseUrl}}${{path}}`;
+    /** @type {{_MemEntry<T> | undefined}} */
+    const memHit = this._memGet(url);
+    const browserCache = this._browserCache;
+
+    // L1 fast path: deliver from memCache, revalidate via network.
+    // ETag match → zero parse, zero clone, zero cache write, no second onValue fire.
+    if (memHit) {{
+      if (onValue) onValue(memHit.value);
+      try {{
+        const res = await this.get(path, {{ signal }});
+        const netEtag = res.headers.get('ETag');
+        if (netEtag && netEtag === memHit.etag) return memHit.value;
+        const cloned = browserCache ? res.clone() : null;
+        const value = await parse(res);
+        this._memSet(url, netEtag, value);
+        if (onValue) onValue(value);
+        if (cloned && browserCache) {{
+          const cacheStore = browserCache;
+          _runIdle(() => cacheStore.put(url, cloned));
+        }}
+        return value;
+      }} catch {{
+        return memHit.value;
+      }}
+    }}
+
+    // L1 miss: race browserCache (stale snapshot) vs network (fresh).
+    let networkSettled = false;
+    const stalePromise = onValue && browserCache
+      ? browserCache.match(url).then(async (res) => {{
+          if (!res || networkSettled) return null;
+          const value = await parse(res);
+          if (networkSettled) return value;
+          this._memSet(url, res.headers.get('ETag'), value);
+          onValue(value);
+          return value;
+        }}).catch(() => null)
+      : null;
+
+    try {{
+      const res = await this.get(path, {{ signal }});
+      networkSettled = true;
+      const netEtag = res.headers.get('ETag');
+      // Stale won and populated memCache with matching ETag → reuse, skip parse + second onValue.
+      const populated = /** @type {{_MemEntry<T> | undefined}} */ (this._memGet(url));
+      if (populated && netEtag && netEtag === populated.etag) return populated.value;
+      const cloned = browserCache ? res.clone() : null;
+      const value = await parse(res);
+      this._memSet(url, netEtag, value);
+      if (onValue) onValue(value);
+      if (cloned && browserCache) {{
+        const cacheStore = browserCache;
+        _runIdle(() => cacheStore.put(url, cloned));
+      }}
+      return value;
+    }} catch (e) {{
+      const stale = await stalePromise;
+      if (stale != null) return stale;
+      throw e;
+    }}
+  }}
+
+  /**
+   * Make a GET request expecting a JSON response. Cached and supports `onValue`.
+   * @template T
+   * @param {{string}} path
+   * @param {{{{ onValue?: (value: T) => void, signal?: AbortSignal, cache?: boolean }}}} [options]
+   * @returns {{Promise<T>}}
+   */
+  getJson(path, options) {{
+    return this._getCached(path, async (res) => _addCamelGetters(await res.json()), options);
+  }}
+
+  /**
+   * Make a GET request expecting a text response (text/plain, text/csv, ...).
+   * Cached and supports `onValue`, same as `getJson`.
+   * @param {{string}} path
+   * @param {{{{ onValue?: (value: string) => void, signal?: AbortSignal, cache?: boolean }}}} [options]
+   * @returns {{Promise<string>}}
+   */
+  getText(path, options) {{
+    return this._getCached(path, (res) => res.text(), options);
+  }}
+
+  /**
+   * Make a GET request expecting binary data (application/octet-stream).
+   * Cached and supports `onValue`, same as `getJson`.
+   * @param {{string}} path
+   * @param {{{{ onValue?: (value: Uint8Array) => void, signal?: AbortSignal, cache?: boolean }}}} [options]
+   * @returns {{Promise<Uint8Array>}}
+   */
+  getBytes(path, options) {{
+    return this._getCached(path, async (res) => new Uint8Array(await res.arrayBuffer()), options);
+  }}
+
+  /**
+   * Make a POST request with a string body.
+   *
+   * POST responses are uncached and never invoke `onValue` — every call hits
+   * the network with the same body and returns the upstream response.
+   *
+   * @param {{string}} path
+   * @param {{string}} body
+   * @param {{{{ signal?: AbortSignal }}}} [options]
+   * @returns {{Promise<Response>}}
+   */
+  async post(path, body, {{ signal }} = {{}}) {{
+    const url = `${{this.baseUrl}}${{path}}`;
+    const signals = [AbortSignal.timeout(this.timeout)];
+    if (signal) signals.push(signal);
+    const res = await fetch(url, {{
+      method: 'POST',
+      body,
+      signal: AbortSignal.any(signals),
+    }});
+    if (!res.ok) throw new BrkError(`HTTP ${{res.status}}: ${{url}}`, res.status);
+    return res;
+  }}
+
+  /**
+   * Make a POST request expecting a JSON response.
+   * @template T
+   * @param {{string}} path
+   * @param {{string}} body
+   * @param {{{{ signal?: AbortSignal }}}} [options]
+   * @returns {{Promise<T>}}
+   */
+  async postJson(path, body, options) {{
+    const res = await this.post(path, body, options);
+    return _addCamelGetters(await res.json());
+  }}
+
+  /**
+   * Make a POST request expecting a text response.
+   * @param {{string}} path
+   * @param {{string}} body
+   * @param {{{{ signal?: AbortSignal }}}} [options]
+   * @returns {{Promise<string>}}
+   */
+  async postText(path, body, options) {{
+    const res = await this.post(path, body, options);
+    return res.text();
+  }}
+
+  /**
+   * Make a POST request expecting binary data (application/octet-stream).
+   * @param {{string}} path
+   * @param {{string}} body
+   * @param {{{{ signal?: AbortSignal }}}} [options]
+   * @returns {{Promise<Uint8Array>}}
+   */
+  async postBytes(path, body, options) {{
+    const res = await this.post(path, body, options);
+    return new Uint8Array(await res.arrayBuffer());
+  }}
+
+  /**
+   * Fetch series data and wrap with helper methods (internal)
+   * @template T
+   * @param {{string}} path
+   * @param {{(value: DateSeriesData<T>) => void}} [onValue]
+   * @returns {{Promise<DateSeriesData<T>>}}
+   */
+  async _fetchSeriesData(path, onValue) {{
+    const wrappedOnValue = onValue ? (/** @type {{SeriesData<T>}} */ raw) => onValue(_wrapSeriesData(raw)) : undefined;
+    const raw = await this.getJson(path, {{ onValue: wrappedOnValue }});
+    return _wrapSeriesData(raw);
+  }}
+}}
+
+/**
+ * Build series name with suffix.
+ * @param {{string}} acc - Accumulated prefix
+ * @param {{string}} s - Series suffix
+ * @returns {{string}}
+ */
+const _m = (acc, s) => s ? (acc ? `${{acc}}_${{s}}` : s) : acc;
+
+/**
+ * Build series name with prefix.
+ * @param {{string}} prefix - Prefix to prepend
+ * @param {{string}} acc - Accumulated name
+ * @returns {{string}}
+ */
+const _p = (prefix, acc) => acc ? `${{prefix}}_${{acc}}` : prefix;
+
+"#
+    )
+    .unwrap();
+}
+
+/// Generate static constants for the BrkClient class.
+pub fn generate_static_constants(output: &mut String) {
+    let constants = ClientConstants::collect();
+
+    // VERSION, INDEXES, POOL_ID_TO_POOL_NAME
+    writeln!(output, "  VERSION = \"{}\";\n", constants.version).unwrap();
+    write_static_const(output, "INDEXES", &format_json(&constants.indexes));
+    write_static_const(
+        output,
+        "POOL_ID_TO_POOL_NAME",
+        &format_json(&constants.pool_map),
+    );
+
+    // Cohort constants with camelCase keys
+    for (name, value) in CohortConstants::all() {
+        write_static_const(output, name, &format_json(&camel_case_keys(value)));
+    }
+
+    // Helper methods
+    writeln!(
+        output,
+        r#"  /**
+   * Convert an index value to a Date for date-based indexes.
+   * @param {{Index}} index - The index type
+   * @param {{number}} i - The index value
+   * @returns {{globalThis.Date}}
+   */
+  indexToDate(index, i) {{
+    return indexToDate(index, i);
+  }}
+
+  /**
+   * Convert a Date to an index value for date-based indexes.
+   * @param {{Index}} index - The index type
+   * @param {{globalThis.Date}} d - The date to convert
+   * @returns {{number}}
+   */
+  dateToIndex(index, d) {{
+    return dateToIndex(index, d);
+  }}
+
+"#
+    )
+    .unwrap();
+}
+
+fn indent_json_const(json: &str) -> String {
+    json.lines()
+        .enumerate()
+        .map(|(i, line)| {
+            if i == 0 {
+                line.to_string()
+            } else {
+                format!("  {}", line)
+            }
+        })
+        .collect::<Vec<_>>()
+        .join("\n")
+}
+
+fn write_static_const(output: &mut String, name: &str, json: &str) {
+    writeln!(
+        output,
+        "  {} = /** @type {{const}} */ ({});\n",
+        name,
+        indent_json_const(json)
+    )
+    .unwrap();
+}
+
+/// Generate index accessor factory functions.
+pub fn generate_index_accessors(output: &mut String, patterns: &[IndexSetPattern]) {
+    if patterns.is_empty() {
+        return;
+    }
+
+    writeln!(output, "// Index group constants and factory\n").unwrap();
+
+    // Generate index array constants (e.g., _i1 = ["day1", "height"])
+    for (i, pattern) in patterns.iter().enumerate() {
+        write!(output, "const _i{} = /** @type {{const}} */ ([", i + 1).unwrap();
+        for (j, index) in pattern.indexes.iter().enumerate() {
+            if j > 0 {
+                write!(output, ", ").unwrap();
+            }
+            write!(output, "\"{}\"", index.name()).unwrap();
+        }
+        writeln!(output, "]);").unwrap();
+    }
+    writeln!(output).unwrap();
+
+    // Generate ONE generic series pattern factory
+    writeln!(
+        output,
+        r#"/**
+ * Generic series pattern factory.
+ * @template T
+ * @param {{BrkClient}} client
+ * @param {{string}} name - The series vec name
+ * @param {{readonly Index[]}} indexes - The supported indexes
+ */
+function _mp(client, name, indexes) {{
+  const by = {{}};
+  for (const idx of indexes) {{
+    Object.defineProperty(by, idx, {{
+      get() {{ return _endpoint(client, name, idx); }},
+      enumerable: true,
+      configurable: true
+    }});
+  }}
+  return {{
+    name,
+    by,
+    /** @returns {{readonly Index[]}} */
+    indexes() {{ return indexes; }},
+    /** @param {{Index}} index @returns {{SeriesEndpoint<T>|undefined}} */
+    get(index) {{ return indexes.includes(index) ? _endpoint(client, name, index) : undefined; }}
+  }};
+}}
+"#
+    )
+    .unwrap();
+
+    // Generate typedefs and thin wrapper functions
+    for (i, pattern) in patterns.iter().enumerate() {
+        // Generate typedef for type safety
+        let by_fields: Vec<String> = pattern
+            .indexes
+            .iter()
+            .map(|idx| {
+                let builder = if idx.is_date_based() {
+                    "DateSeriesEndpoint"
+                } else {
+                    "SeriesEndpoint"
+                };
+                format!("readonly {}: {}<T>", idx.name(), builder)
+            })
+            .collect();
+        let by_type = format!("{{ {} }}", by_fields.join(", "));
+
+        writeln!(
+            output,
+            "/** @template T @typedef {{{{ name: string, by: {}, indexes: () => readonly Index[], get: (index: Index) => SeriesEndpoint<T>|undefined }}}} {} */",
+            by_type, pattern.name
+        )
+        .unwrap();
+
+        // Generate thin wrapper that calls the generic factory
+        writeln!(
+            output,
+            "/** @template T @param {{BrkClient}} client @param {{string}} name @returns {{{}<T>}} */",
+            pattern.name
+        )
+        .unwrap();
+        writeln!(
+            output,
+            "function create{}(client, name) {{ return /** @type {{{}<T>}} */ (_mp(client, name, _i{})); }}",
+            pattern.name,
+            pattern.name,
+            i + 1
+        )
+        .unwrap();
+    }
+    writeln!(output).unwrap();
+}
+
+/// Generate structural pattern factory functions.
+pub fn generate_structural_patterns(
+    output: &mut String,
+    patterns: &[StructuralPattern],
+    metadata: &ClientMetadata,
+) {
+    if patterns.is_empty() {
+        return;
+    }
+
+    writeln!(output, "// Reusable structural pattern factories\n").unwrap();
+
+    for pattern in patterns {
+        // Generate typedef
+        writeln!(output, "/**").unwrap();
+        if pattern.is_generic {
+            writeln!(output, " * @template T").unwrap();
+        }
+        writeln!(output, " * @typedef {{Object}} {}", pattern.name).unwrap();
+        for field in &pattern.fields {
+            let js_type = metadata.field_type_annotation(
+                field,
+                pattern.is_generic,
+                None,
+                GenericSyntax::JAVASCRIPT,
+            );
+            writeln!(
+                output,
+                " * @property {{{}}} {}",
+                js_type,
+                to_camel_case(&field.name)
+            )
+            .unwrap();
+        }
+        writeln!(output, " */\n").unwrap();
+
+        // Skip factory for non-parameterizable patterns (inlined at tree level)
+        if !metadata.is_parameterizable(&pattern.name) {
+            continue;
+        }
+
+        writeln!(output, "/**").unwrap();
+        writeln!(output, " * Create a {} pattern node", pattern.name).unwrap();
+        if pattern.is_generic {
+            writeln!(output, " * @template T").unwrap();
+        }
+        writeln!(output, " * @param {{BrkClient}} client").unwrap();
+        writeln!(output, " * @param {{string}} acc - Accumulated series name").unwrap();
+        if pattern.is_templated() {
+            writeln!(output, " * @param {{string}} disc - Discriminator suffix").unwrap();
+        }
+        let return_type = if pattern.is_generic {
+            format!("{}<T>", pattern.name)
+        } else {
+            pattern.name.clone()
+        };
+        writeln!(output, " * @returns {{{}}}", return_type).unwrap();
+        writeln!(output, " */").unwrap();
+
+        if pattern.is_templated() {
+            writeln!(
+                output,
+                "function create{}(client, acc, disc) {{",
+                pattern.name
+            )
+            .unwrap();
+        } else {
+            writeln!(output, "function create{}(client, acc) {{", pattern.name).unwrap();
+        }
+        writeln!(output, "  return {{").unwrap();
+
+        let syntax = JavaScriptSyntax;
+        for field in &pattern.fields {
+            generate_parameterized_field(output, &syntax, field, pattern, metadata, "    ");
+        }
+
+        writeln!(output, "  }};").unwrap();
+        writeln!(output, "}}\n").unwrap();
+    }
+}

crates/brk_bindgen/src/generators/javascript/mod.rs

@@ -0,0 +1,66 @@
+//! JavaScript client generation.
+//!
+//! This module generates a JavaScript + JSDoc client for the BRK API.
+
+mod api;
+pub mod client;
+pub mod tree;
+pub mod types;
+
+use std::{fmt::Write, fs, io, path::Path};
+
+use serde_json::json;
+
+use super::write_if_changed;
+use crate::{ClientMetadata, Endpoint, TypeSchemas, VERSION};
+
+/// Generate JavaScript + JSDoc client from metadata and OpenAPI endpoints.
+///
+/// `output_path` is the full path to the output file (e.g., "modules/brk-client/index.js").
+pub fn generate_javascript_client(
+    metadata: &ClientMetadata,
+    endpoints: &[Endpoint],
+    schemas: &TypeSchemas,
+    output_path: &Path,
+) -> io::Result<()> {
+    let mut output = String::new();
+
+    writeln!(output, "// Auto-generated BRK JavaScript client").unwrap();
+    writeln!(output, "// Do not edit manually\n").unwrap();
+
+    types::generate_type_definitions(&mut output, schemas);
+    client::generate_base_client(&mut output);
+    client::generate_index_accessors(&mut output, &metadata.index_set_patterns);
+    client::generate_structural_patterns(&mut output, &metadata.structural_patterns, metadata);
+    tree::generate_tree_typedefs(&mut output, &metadata.catalog, metadata);
+    tree::generate_main_client(&mut output, &metadata.catalog, metadata, endpoints);
+
+    write_if_changed(output_path, &output)?;
+
+    // Update package.json version if it exists in the same directory
+    if let Some(parent) = output_path.parent() {
+        let package_json_path = parent.join("package.json");
+        if package_json_path.exists() {
+            update_package_json_version(&package_json_path)?;
+        }
+    }
+
+    Ok(())
+}
+
+fn update_package_json_version(package_json_path: &Path) -> io::Result<()> {
+    let content = fs::read_to_string(package_json_path)?;
+    let mut package: serde_json::Value = serde_json::from_str(&content)
+        .map_err(|e| io::Error::new(io::ErrorKind::InvalidData, e))?;
+
+    if let Some(obj) = package.as_object_mut() {
+        obj.insert("version".to_string(), json!(VERSION));
+    }
+
+    let updated = serde_json::to_string_pretty(&package)
+        .map_err(|e| io::Error::new(io::ErrorKind::InvalidData, e))?;
+
+    write_if_changed(package_json_path, &(updated + "\n"))?;
+
+    Ok(())
+}

crates/brk_bindgen/src/generators/javascript/tree.rs

@@ -0,0 +1,227 @@
+//! JavaScript tree structure generation.
+
+use std::collections::BTreeSet;
+use std::fmt::Write;
+
+use brk_types::TreeNode;
+
+use crate::{
+    ClientMetadata, Endpoint, GenericSyntax, JavaScriptSyntax, PatternField, build_child_path,
+    generate_leaf_field, generate_tree_node_field, prepare_tree_node, to_camel_case,
+};
+
+use super::api::generate_api_methods;
+use super::client::generate_static_constants;
+
+/// Generate JSDoc typedefs for the series tree.
+pub fn generate_tree_typedefs(output: &mut String, catalog: &TreeNode, metadata: &ClientMetadata) {
+    writeln!(output, "// Catalog tree typedefs\n").unwrap();
+
+    let pattern_lookup = metadata.pattern_lookup();
+    let mut generated = BTreeSet::new();
+    generate_tree_typedef(
+        output,
+        "SeriesTree",
+        "",
+        catalog,
+        pattern_lookup,
+        metadata,
+        &mut generated,
+    );
+}
+
+fn generate_tree_typedef(
+    output: &mut String,
+    name: &str,
+    path: &str,
+    node: &TreeNode,
+    pattern_lookup: &std::collections::BTreeMap<Vec<PatternField>, String>,
+    metadata: &ClientMetadata,
+    generated: &mut BTreeSet<String>,
+) {
+    let Some(ctx) = prepare_tree_node(node, name, path, pattern_lookup, metadata, generated) else {
+        return;
+    };
+
+    writeln!(output, "/**").unwrap();
+    writeln!(output, " * @typedef {{Object}} {}", name).unwrap();
+
+    for child in &ctx.children {
+        let js_type = if child.should_inline {
+            child.inline_type_name.clone()
+        } else {
+            metadata.field_type_annotation(&child.field, false, None, GenericSyntax::JAVASCRIPT)
+        };
+
+        writeln!(
+            output,
+            " * @property {{{}}} {}",
+            js_type,
+            to_camel_case(&child.field.name)
+        )
+        .unwrap();
+    }
+
+    writeln!(output, " */\n").unwrap();
+
+    // Generate child typedefs
+    for child in &ctx.children {
+        if child.should_inline {
+            let child_path = build_child_path(path, child.name);
+            generate_tree_typedef(
+                output,
+                &child.inline_type_name,
+                &child_path,
+                child.node,
+                pattern_lookup,
+                metadata,
+                generated,
+            );
+        }
+    }
+}
+
+/// Generate the main BrkClient class.
+pub fn generate_main_client(
+    output: &mut String,
+    catalog: &TreeNode,
+    metadata: &ClientMetadata,
+    endpoints: &[Endpoint],
+) {
+    let pattern_lookup = metadata.pattern_lookup();
+
+    writeln!(output, "/**").unwrap();
+    writeln!(
+        output,
+        " * Main BRK client with series tree and API methods"
+    )
+    .unwrap();
+    writeln!(output, " * @extends BrkClientBase").unwrap();
+    writeln!(output, " */").unwrap();
+    writeln!(output, "class BrkClient extends BrkClientBase {{").unwrap();
+
+    generate_static_constants(output);
+
+    writeln!(output, "  /**").unwrap();
+    writeln!(output, "   * @param {{BrkClientOptions|string}} options").unwrap();
+    writeln!(output, "   */").unwrap();
+    writeln!(output, "  constructor(options) {{").unwrap();
+    writeln!(output, "    super(options);").unwrap();
+    writeln!(output, "    /** @type {{SeriesTree}} */").unwrap();
+    writeln!(output, "    this.series = this._buildTree();").unwrap();
+    writeln!(output, "  }}\n").unwrap();
+
+    writeln!(output, "  /**").unwrap();
+    writeln!(output, "   * @private").unwrap();
+    writeln!(output, "   * @returns {{SeriesTree}}").unwrap();
+    writeln!(output, "   */").unwrap();
+    writeln!(output, "  _buildTree() {{").unwrap();
+    writeln!(output, "    return {{").unwrap();
+    let mut generated = BTreeSet::new();
+    generate_tree_initializer(
+        output,
+        catalog,
+        "SeriesTree",
+        "",
+        3,
+        pattern_lookup,
+        metadata,
+        &mut generated,
+    );
+    writeln!(output, "    }};").unwrap();
+    writeln!(output, "  }}\n").unwrap();
+
+    writeln!(output, "  /**").unwrap();
+    writeln!(
+        output,
+        "   * Create a dynamic series endpoint builder for any series/index combination."
+    )
+    .unwrap();
+    writeln!(output, "   *").unwrap();
+    writeln!(
+        output,
+        "   * Use this for programmatic access when the series name is determined at runtime."
+    )
+    .unwrap();
+    writeln!(
+        output,
+        "   * For type-safe access, use the `series` tree instead."
+    )
+    .unwrap();
+    writeln!(output, "   *").unwrap();
+    writeln!(output, "   * @param {{string}} series - The series name").unwrap();
+    writeln!(output, "   * @param {{Index}} index - The index name").unwrap();
+    writeln!(output, "   * @returns {{SeriesEndpoint<unknown>}}").unwrap();
+    writeln!(output, "   */").unwrap();
+    writeln!(output, "  seriesEndpoint(series, index) {{").unwrap();
+    writeln!(output, "    return _endpoint(this, series, index);").unwrap();
+    writeln!(output, "  }}\n").unwrap();
+
+    generate_api_methods(output, endpoints);
+
+    writeln!(output, "}}\n").unwrap();
+
+    writeln!(output, "export {{ BrkClient, BrkError }};").unwrap();
+}
+
+#[allow(clippy::too_many_arguments)]
+fn generate_tree_initializer(
+    output: &mut String,
+    node: &TreeNode,
+    name: &str,
+    path: &str,
+    indent: usize,
+    pattern_lookup: &std::collections::BTreeMap<Vec<PatternField>, String>,
+    metadata: &ClientMetadata,
+    generated: &mut BTreeSet<String>,
+) {
+    let indent_str = "  ".repeat(indent);
+
+    let Some(ctx) = prepare_tree_node(node, name, path, pattern_lookup, metadata, generated) else {
+        return;
+    };
+
+    let syntax = JavaScriptSyntax;
+    for child in &ctx.children {
+        let field_name = to_camel_case(child.name);
+
+        if child.is_leaf {
+            if let TreeNode::Leaf(leaf) = child.node {
+                generate_leaf_field(
+                    output,
+                    &syntax,
+                    "this",
+                    child.name,
+                    leaf,
+                    metadata,
+                    &indent_str,
+                );
+            }
+        } else if child.should_inline {
+            // Inline object
+            let child_path = build_child_path(path, child.name);
+            writeln!(output, "{}{}: {{", indent_str, field_name).unwrap();
+            generate_tree_initializer(
+                output,
+                child.node,
+                &child.inline_type_name,
+                &child_path,
+                indent + 1,
+                pattern_lookup,
+                metadata,
+                generated,
+            );
+            writeln!(output, "{}}},", indent_str).unwrap();
+        } else {
+            generate_tree_node_field(
+                output,
+                &syntax,
+                &child.field,
+                metadata,
+                &indent_str,
+                "this",
+                &child.base_result,
+            );
+        }
+    }
+}

crates/brk_bindgen/src/generators/javascript/types.rs

@@ -0,0 +1,220 @@
+//! JavaScript type definitions generation.
+
+use std::fmt::Write;
+
+use serde_json::Value;
+
+use crate::{
+    TypeSchemas,
+    generators::{MANUAL_GENERIC_TYPES, write_description},
+    get_union_variants, ref_to_type_name, to_camel_case,
+};
+
+/// Generate JSDoc type definitions from OpenAPI schemas.
+pub fn generate_type_definitions(output: &mut String, schemas: &TypeSchemas) {
+    if schemas.is_empty() {
+        return;
+    }
+
+    writeln!(output, "// Type definitions\n").unwrap();
+
+    for (name, schema) in schemas {
+        if MANUAL_GENERIC_TYPES.contains(&name.as_str()) {
+            continue;
+        }
+
+        let js_type = schema_to_js_type(schema, Some(name));
+
+        let type_desc = schema.get("description").and_then(|d| d.as_str());
+
+        if is_primitive_alias(schema) {
+            if let Some(desc) = type_desc {
+                writeln!(output, "/**").unwrap();
+                write_description(output, desc, " * ", " *");
+                writeln!(output, " *").unwrap();
+                writeln!(output, " * @typedef {{{}}} {}", js_type, name).unwrap();
+                writeln!(output, " */").unwrap();
+            } else {
+                writeln!(output, "/** @typedef {{{}}} {} */", js_type, name).unwrap();
+            }
+        } else if let Some(props) = schema.get("properties").and_then(|p| p.as_object()) {
+            writeln!(output, "/**").unwrap();
+            if let Some(desc) = type_desc {
+                write_description(output, desc, " * ", " *");
+                writeln!(output, " *").unwrap();
+            }
+            writeln!(output, " * @typedef {{Object}} {}", name).unwrap();
+            for (prop_name, prop_schema) in props {
+                let prop_type = schema_to_js_type(prop_schema, Some(name));
+                let required = schema
+                    .get("required")
+                    .and_then(|r| r.as_array())
+                    .map(|arr| arr.iter().any(|v| v.as_str() == Some(prop_name)))
+                    .unwrap_or(false);
+                let optional = if required { "" } else { "=" };
+                let safe_name = to_camel_case(&prop_name.replace(['[', ']'], ""));
+                let prop_desc = prop_schema
+                    .get("description")
+                    .and_then(|d| d.as_str())
+                    .map(|d| format!(" - {}", d))
+                    .unwrap_or_default();
+                writeln!(
+                    output,
+                    " * @property {{{}{}}} {}{}",
+                    prop_type, optional, safe_name, prop_desc
+                )
+                .unwrap();
+            }
+            writeln!(output, " */").unwrap();
+        } else if let Some(desc) = type_desc {
+            writeln!(output, "/**").unwrap();
+            write_description(output, desc, " * ", " *");
+            writeln!(output, " *").unwrap();
+            writeln!(output, " * @typedef {{{}}} {}", js_type, name).unwrap();
+            writeln!(output, " */").unwrap();
+        } else {
+            writeln!(output, "/** @typedef {{{}}} {} */", js_type, name).unwrap();
+        }
+    }
+    writeln!(output).unwrap();
+}
+
+fn is_primitive_alias(schema: &Value) -> bool {
+    schema.get("properties").is_none()
+        && schema.get("items").is_none()
+        && schema.get("anyOf").is_none()
+        && schema.get("oneOf").is_none()
+        && schema.get("enum").is_none()
+}
+
+fn json_type_to_js(ty: &str, schema: &Value, current_type: Option<&str>) -> String {
+    match ty {
+        "integer" | "number" => "number".to_string(),
+        "boolean" => "boolean".to_string(),
+        "string" => "string".to_string(),
+        "null" => "null".to_string(),
+        "array" => {
+            let item_type = schema
+                .get("items")
+                .map(|s| schema_to_js_type(s, current_type))
+                .unwrap_or_else(|| "*".to_string());
+            format!("{}[]", item_type)
+        }
+        "object" => {
+            if let Some(add_props) = schema.get("additionalProperties") {
+                let value_type = schema_to_js_type(add_props, current_type);
+                return format!("{{ [key: string]: {} }}", value_type);
+            }
+            "Object".to_string()
+        }
+        _ => "*".to_string(),
+    }
+}
+
+/// JSDoc has no `integer` keyword, only `number`. Map `integer` (and `integer[]`,
+/// `Foo<integer>`, etc.) to `number` before emitting type strings to JS.
+pub fn jsdoc_normalize(ty: &str) -> String {
+    let mut out = ty.to_string();
+    let mut prev = String::new();
+    while prev != out {
+        prev = out.clone();
+        out = out.replace("integer[]", "number[]");
+        out = out.replace("<integer>", "<number>");
+        out = out.replace("(integer)", "(number)");
+        out = out.replace("integer | ", "number | ");
+        out = out.replace(" | integer", " | number");
+    }
+    if out == "integer" {
+        return "number".to_string();
+    }
+    out
+}
+
+/// Convert a JSON schema to a JavaScript type string.
+pub fn schema_to_js_type(schema: &Value, current_type: Option<&str>) -> String {
+    if let Some(all_of) = schema.get("allOf").and_then(|v| v.as_array()) {
+        for item in all_of {
+            let resolved = schema_to_js_type(item, current_type);
+            if resolved != "*" {
+                return resolved;
+            }
+        }
+    }
+
+    if let Some(ref_path) = schema.get("$ref").and_then(|r| r.as_str()) {
+        return ref_to_type_name(ref_path).unwrap_or("*").to_string();
+    }
+
+    if let Some(enum_values) = schema.get("enum").and_then(|e| e.as_array()) {
+        let literals: Vec<String> = enum_values
+            .iter()
+            .filter_map(|v| v.as_str())
+            .map(|s| format!("\"{}\"", s))
+            .collect();
+        if !literals.is_empty() {
+            return format!("({})", literals.join("|"));
+        }
+    }
+
+    if let Some(ty) = schema.get("type") {
+        if let Some(type_array) = ty.as_array() {
+            let types: Vec<String> = type_array
+                .iter()
+                .filter_map(|t| t.as_str())
+                .filter(|t| *t != "null")
+                .map(|t| json_type_to_js(t, schema, current_type))
+                .collect();
+            let has_null = type_array.iter().any(|t| t.as_str() == Some("null"));
+
+            if types.len() == 1 {
+                let base_type = &types[0];
+                return if has_null {
+                    format!("?{}", base_type)
+                } else {
+                    base_type.clone()
+                };
+            } else if !types.is_empty() {
+                let union = format!("({})", types.join("|"));
+                return if has_null {
+                    format!("?{}", union)
+                } else {
+                    union
+                };
+            }
+        }
+
+        if let Some(ty_str) = ty.as_str() {
+            return json_type_to_js(ty_str, schema, current_type);
+        }
+    }
+
+    if let Some(variants) = get_union_variants(schema) {
+        let types: Vec<String> = variants
+            .iter()
+            .map(|v| schema_to_js_type(v, current_type))
+            .collect();
+        let filtered: Vec<_> = types.iter().filter(|t| *t != "*").collect();
+        if !filtered.is_empty() {
+            return format!(
+                "({})",
+                filtered
+                    .iter()
+                    .map(|s| s.as_str())
+                    .collect::<Vec<_>>()
+                    .join("|")
+            );
+        }
+        return format!("({})", types.join("|"));
+    }
+
+    if let Some(format) = schema.get("format").and_then(|f| f.as_str()) {
+        return match format {
+            "int32" | "int64" => "number".to_string(),
+            "float" | "double" => "number".to_string(),
+            "date" | "date-time" => "string".to_string(),
+            _ => "*".to_string(),
+        };
+    }
+
+    "*".to_string()
+}

crates/brk_bindgen/src/generators/mod.rs

@@ -0,0 +1,54 @@
+//! Code generators for client libraries.
+//!
+//! Each language has its own submodule with focused files:
+//! - `types.rs` - Type definitions
+//! - `client.rs` - Base client and pattern factories
+//! - `tree.rs` - Tree structure generation
+//! - `api.rs` - API method generation
+//! - `mod.rs` - Entry point
+
+use std::{fmt::Write, fs, io, path::Path};
+
+pub mod javascript;
+pub mod python;
+pub mod rust;
+
+pub use javascript::generate_javascript_client;
+pub use python::generate_python_client;
+pub use rust::generate_rust_client;
+
+/// Types that are manually defined as generics in client code, not from schema.
+pub const MANUAL_GENERIC_TYPES: &[&str] = &["SeriesData", "SeriesEndpoint"];
+
+/// Write a multi-line description with the given prefix for each line.
+/// `empty_prefix` is used for blank lines (e.g., "   *" without trailing space).
+pub fn write_description(output: &mut String, desc: &str, prefix: &str, empty_prefix: &str) {
+    for line in desc.lines() {
+        if line.is_empty() {
+            writeln!(output, "{}", empty_prefix).unwrap();
+        } else {
+            writeln!(output, "{}{}", prefix, line).unwrap();
+        }
+    }
+}
+
+/// Replace generic types with their Any variants in return types.
+/// Used by JS and Python generators.
+pub fn normalize_return_type(return_type: &str) -> String {
+    let mut result = return_type.to_string();
+    for type_name in MANUAL_GENERIC_TYPES {
+        result = result.replace(type_name, &format!("Any{}", type_name));
+    }
+    result
+}
+
+/// Write content to a file only if it differs from existing content.
+/// Preserves mtime when unchanged, avoiding unnecessary cargo rebuilds.
+pub fn write_if_changed(path: &Path, content: &str) -> io::Result<()> {
+    if let Ok(existing) = fs::read_to_string(path)
+        && existing == content
+    {
+        return Ok(());
+    }
+    fs::write(path, content)
+}

crates/brk_bindgen/src/generators/python/api.rs

@@ -0,0 +1,311 @@
+//! Python API method generation.
+
+use std::fmt::Write;
+
+use crate::{
+    Endpoint, Parameter, escape_python_keyword,
+    generators::{normalize_return_type, write_description},
+    to_snake_case,
+};
+
+use super::client::generate_class_constants;
+use super::types::js_type_to_python;
+
+/// Generate the main client class
+pub fn generate_main_client(output: &mut String, endpoints: &[Endpoint]) {
+    writeln!(output, "class BrkClient(BrkClientBase):").unwrap();
+    writeln!(
+        output,
+        "    \"\"\"Main BRK client with series tree and API methods.\"\"\""
+    )
+    .unwrap();
+    writeln!(output).unwrap();
+
+    // Generate class-level constants
+    generate_class_constants(output);
+
+    writeln!(
+        output,
+        "    def __init__(self, base_url: str = 'http://localhost:3000', timeout: float = 30.0):"
+    )
+    .unwrap();
+    writeln!(output, "        super().__init__(base_url, timeout)").unwrap();
+    writeln!(output, "        self.series = SeriesTree(self)").unwrap();
+    writeln!(output).unwrap();
+
+    // Generate series_endpoint() method for dynamic series access
+    writeln!(
+        output,
+        "    def series_endpoint(self, series: str, index: Index) -> SeriesEndpoint[Any]:"
+    )
+    .unwrap();
+    writeln!(
+        output,
+        "        \"\"\"Create a dynamic series endpoint builder for any series/index combination."
+    )
+    .unwrap();
+    writeln!(output).unwrap();
+    writeln!(
+        output,
+        "        Use this for programmatic access when the series name is determined at runtime."
+    )
+    .unwrap();
+    writeln!(
+        output,
+        "        For type-safe access, use the `series` tree instead."
+    )
+    .unwrap();
+    writeln!(output, "        \"\"\"").unwrap();
+    writeln!(output, "        return SeriesEndpoint(self, series, index)").unwrap();
+    writeln!(output).unwrap();
+
+    // Generate helper methods
+    writeln!(
+        output,
+        "    def index_to_date(self, index: Index, i: int) -> Union[date, datetime]:"
+    )
+    .unwrap();
+    writeln!(
+        output,
+        "        \"\"\"Convert an index value to a date/datetime for date-based indexes.\"\"\""
+    )
+    .unwrap();
+    writeln!(output, "        return _index_to_date(index, i)").unwrap();
+    writeln!(output).unwrap();
+    writeln!(
+        output,
+        "    def date_to_index(self, index: Index, d: Union[date, datetime]) -> int:"
+    )
+    .unwrap();
+    writeln!(
+        output,
+        "        \"\"\"Convert a date/datetime to an index value for date-based indexes.\"\"\""
+    )
+    .unwrap();
+    writeln!(output, "        return _date_to_index(index, d)").unwrap();
+    writeln!(output).unwrap();
+    // Generate API methods
+    generate_api_methods(output, endpoints);
+}
+
+/// Generate API methods from OpenAPI endpoints
+pub fn generate_api_methods(output: &mut String, endpoints: &[Endpoint]) {
+    for endpoint in endpoints {
+        if !endpoint.should_generate() {
+            continue;
+        }
+
+        let method_name = endpoint_to_method_name(endpoint);
+        let base_return_type = if endpoint.returns_binary() {
+            "bytes".to_string()
+        } else {
+            normalize_return_type(
+                &endpoint
+                    .schema_name()
+                    .map(js_type_to_python)
+                    .unwrap_or_else(|| "str".to_string()),
+            )
+        };
+
+        let return_type = if endpoint.supports_csv {
+            format!("Union[{}, str]", base_return_type)
+        } else {
+            base_return_type
+        };
+
+        // Build method signature
+        let params = build_method_params(endpoint);
+        writeln!(
+            output,
+            "    def {}(self{}) -> {}:",
+            method_name, params, return_type
+        )
+        .unwrap();
+
+        // Docstring
+        match (&endpoint.summary, &endpoint.description) {
+            (Some(summary), Some(desc)) if summary != desc => {
+                writeln!(output, "        \"\"\"{}.", summary.trim_end_matches('.')).unwrap();
+                writeln!(output).unwrap();
+                write_description(output, desc, "        ", "");
+            }
+            (Some(summary), _) => {
+                writeln!(output, "        \"\"\"{}", summary).unwrap();
+            }
+            (None, Some(desc)) => {
+                // First line includes opening quotes
+                let mut lines = desc.lines();
+                if let Some(first) = lines.next() {
+                    writeln!(output, "        \"\"\"{}", first).unwrap();
+                }
+                for line in lines {
+                    if line.is_empty() {
+                        writeln!(output).unwrap();
+                    } else {
+                        writeln!(output, "        {}", line).unwrap();
+                    }
+                }
+            }
+            (None, None) => {
+                write!(output, "        \"\"\"").unwrap();
+            }
+        }
+        writeln!(output).unwrap();
+        writeln!(
+            output,
+            "        Endpoint: `{} {}`\"\"\"",
+            endpoint.method.to_uppercase(),
+            endpoint.path
+        )
+        .unwrap();
+
+        // Build path
+        let path = build_path_template(&endpoint.path, &endpoint.path_params);
+
+        let is_post = endpoint.method == "POST";
+        let fetch_method = match (is_post, &endpoint.response_kind) {
+            (false, _) if endpoint.returns_binary() => "get",
+            (false, _) if endpoint.returns_json() => "get_json",
+            (false, _) => "get_text",
+            (true, _) if endpoint.returns_binary() => "post",
+            (true, _) if endpoint.returns_json() => "post_json",
+            (true, _) => "post_text",
+        };
+
+        let body_arg = if is_post && endpoint.request_body.is_some() {
+            ", body"
+        } else {
+            ""
+        };
+
+        let (wrap_prefix, wrap_suffix) = if endpoint.response_kind.text_is_numeric() {
+            ("int(", ")")
+        } else {
+            ("", "")
+        };
+
+        if endpoint.query_params.is_empty() {
+            if endpoint.path_params.is_empty() {
+                writeln!(
+                    output,
+                    "        return {}self.{}('{}'{}){}",
+                    wrap_prefix, fetch_method, path, body_arg, wrap_suffix
+                )
+                .unwrap();
+            } else {
+                writeln!(
+                    output,
+                    "        return {}self.{}(f'{}'{}){}",
+                    wrap_prefix, fetch_method, path, body_arg, wrap_suffix
+                )
+                .unwrap();
+            }
+        } else {
+            writeln!(output, "        params = []").unwrap();
+            for param in &endpoint.query_params {
+                // Use safe name for Python variable, original name for API query parameter
+                let safe_name = escape_python_keyword(&param.name);
+                let is_array = param.param_type.ends_with("[]");
+                if is_array {
+                    writeln!(
+                        output,
+                        "        for _v in {}: params.append(f'{}={{_v}}')",
+                        safe_name, param.name
+                    )
+                    .unwrap();
+                } else if param.required {
+                    writeln!(
+                        output,
+                        "        params.append(f'{}={{{}}}')",
+                        param.name, safe_name
+                    )
+                    .unwrap();
+                } else {
+                    writeln!(
+                        output,
+                        "        if {} is not None: params.append(f'{}={{{}}}')",
+                        safe_name, param.name, safe_name
+                    )
+                    .unwrap();
+                }
+            }
+            writeln!(output, "        query = '&'.join(params)").unwrap();
+            writeln!(
+                output,
+                "        path = f'{}{{\"?\" + query if query else \"\"}}'",
+                path
+            )
+            .unwrap();
+
+            if endpoint.supports_csv {
+                writeln!(output, "        if format == 'csv':").unwrap();
+                writeln!(output, "            return self.get_text(path)").unwrap();
+                writeln!(
+                    output,
+                    "        return {}self.{}(path{}){}",
+                    wrap_prefix, fetch_method, body_arg, wrap_suffix
+                )
+                .unwrap();
+            } else {
+                writeln!(
+                    output,
+                    "        return {}self.{}(path{}){}",
+                    wrap_prefix, fetch_method, body_arg, wrap_suffix
+                )
+                .unwrap();
+            }
+        }
+
+        writeln!(output).unwrap();
+    }
+}
+
+fn endpoint_to_method_name(endpoint: &Endpoint) -> String {
+    to_snake_case(&endpoint.operation_name())
+}
+
+fn build_method_params(endpoint: &Endpoint) -> String {
+    let mut params = Vec::new();
+    // Path params are always required
+    for param in &endpoint.path_params {
+        let safe_name = escape_python_keyword(&param.name);
+        let py_type = js_type_to_python(&param.param_type);
+        params.push(format!(", {}: {}", safe_name, py_type));
+    }
+    // Required query params must come before optional ones (Python syntax requirement)
+    for param in &endpoint.query_params {
+        if param.required {
+            let safe_name = escape_python_keyword(&param.name);
+            let py_type = js_type_to_python(&param.param_type);
+            params.push(format!(", {}: {}", safe_name, py_type));
+        }
+    }
+    for param in &endpoint.query_params {
+        if !param.required {
+            let safe_name = escape_python_keyword(&param.name);
+            let py_type = js_type_to_python(&param.param_type);
+            params.push(format!(", {}: Optional[{}] = None", safe_name, py_type));
+        }
+    }
+    if let Some(body) = &endpoint.request_body {
+        let py_type = js_type_to_python(&body.body_type);
+        if body.required {
+            params.push(format!(", body: {}", py_type));
+        } else {
+            params.push(format!(", body: Optional[{}] = None", py_type));
+        }
+    }
+    params.join("")
+}
+
+fn build_path_template(path: &str, path_params: &[Parameter]) -> String {
+    let mut result = path.to_string();
+    for param in path_params {
+        let placeholder = format!("{{{}}}", param.name);
+        // Use escaped name for Python variable interpolation in f-string
+        let safe_name = escape_python_keyword(&param.name);
+        let interpolation = format!("{{{}}}", safe_name);
+        result = result.replace(&placeholder, &interpolation);
+    }
+    result
+}

crates/brk_bindgen/src/generators/python/client.rs

@@ -0,0 +1,775 @@
+//! Python base client and pattern factory generation.
+
+use std::fmt::Write;
+
+use crate::{
+    ClientConstants, ClientMetadata, CohortConstants, IndexSetPattern, PythonSyntax,
+    StructuralPattern, format_json, generate_parameterized_field, index_to_field_name,
+};
+
+/// Generate class-level constants for the BrkClient class.
+pub fn generate_class_constants(output: &mut String) {
+    let constants = ClientConstants::collect();
+
+    // VERSION
+    writeln!(output, "    VERSION = \"{}\"\n", constants.version).unwrap();
+
+    // INDEXES, POOL_ID_TO_POOL_NAME
+    write_class_const(output, "INDEXES", &format_json(&constants.indexes));
+    // Python needs string keys for pool map
+    let pool_map: std::collections::BTreeMap<String, &str> = constants
+        .pool_map
+        .iter()
+        .map(|(k, v)| (k.to_string(), *v))
+        .collect();
+    write_class_const(output, "POOL_ID_TO_POOL_NAME", &format_json(&pool_map));
+
+    // Cohort constants (no camelCase conversion for Python)
+    for (name, value) in CohortConstants::all() {
+        write_class_const(output, name, &format_json(&value));
+    }
+}
+
+fn write_class_const(output: &mut String, name: &str, json: &str) {
+    let indented = json
+        .lines()
+        .enumerate()
+        .map(|(i, line)| {
+            if i == 0 {
+                format!("    {} = {}", name, line)
+            } else {
+                format!("    {}", line)
+            }
+        })
+        .collect::<Vec<_>>()
+        .join("\n");
+    writeln!(output, "{}\n", indented).unwrap();
+}
+
+/// Generate the base BrkClient class with HTTP functionality
+pub fn generate_base_client(output: &mut String) {
+    writeln!(
+        output,
+        r#"class BrkError(Exception):
+    """Custom error class for BRK client errors."""
+
+    def __init__(self, message: str, status: Optional[int] = None):
+        super().__init__(message)
+        self.status = status
+
+
+class BrkClientBase:
+    """Base HTTP client for making requests."""
+
+    def __init__(self, base_url: str, timeout: float = 30.0):
+        parsed = urlparse(base_url)
+        self._host = parsed.netloc
+        self._secure = parsed.scheme == 'https'
+        self._timeout = timeout
+        self._conn: Optional[Union[HTTPSConnection, HTTPConnection]] = None
+
+    def _connect(self) -> Union[HTTPSConnection, HTTPConnection]:
+        """Get or create HTTP connection."""
+        if self._conn is None:
+            if self._secure:
+                self._conn = HTTPSConnection(self._host, timeout=self._timeout)
+            else:
+                self._conn = HTTPConnection(self._host, timeout=self._timeout)
+        return self._conn
+
+    def get(self, path: str) -> bytes:
+        """Make a GET request and return raw bytes."""
+        try:
+            conn = self._connect()
+            conn.request("GET", path)
+            res = conn.getresponse()
+            data = res.read()
+            if res.status >= 400:
+                raise BrkError(f"HTTP error: {{res.status}}", res.status)
+            return data
+        except (ConnectionError, OSError, TimeoutError) as e:
+            self._conn = None
+            raise BrkError(str(e))
+
+    def get_json(self, path: str) -> Any:
+        """Make a GET request and return JSON."""
+        return json.loads(self.get(path))
+
+    def get_text(self, path: str) -> str:
+        """Make a GET request and return text."""
+        return self.get(path).decode()
+
+    def post(self, path: str, body: str) -> bytes:
+        """Make a POST request with a string body and return raw bytes."""
+        try:
+            conn = self._connect()
+            conn.request("POST", path, body=body)
+            res = conn.getresponse()
+            data = res.read()
+            if res.status >= 400:
+                raise BrkError(f"HTTP error: {{res.status}}", res.status)
+            return data
+        except (ConnectionError, OSError, TimeoutError) as e:
+            self._conn = None
+            raise BrkError(str(e))
+
+    def post_json(self, path: str, body: str) -> Any:
+        """Make a POST request and return JSON."""
+        return json.loads(self.post(path, body))
+
+    def post_text(self, path: str, body: str) -> str:
+        """Make a POST request and return text."""
+        return self.post(path, body).decode()
+
+    def close(self) -> None:
+        """Close the HTTP client."""
+        if self._conn:
+            self._conn.close()
+            self._conn = None
+
+    def __enter__(self) -> BrkClientBase:
+        return self
+
+    def __exit__(self, exc_type: Optional[type], exc_val: Optional[BaseException], exc_tb: Optional[Any]) -> None:
+        self.close()
+
+
+def _m(acc: str, s: str) -> str:
+    """Build series name with suffix."""
+    if not s: return acc
+    return f"{{acc}}_{{s}}" if acc else s
+
+
+def _p(prefix: str, acc: str) -> str:
+    """Build series name with prefix."""
+    return f"{{prefix}}_{{acc}}" if acc else prefix
+
+"#
+    )
+    .unwrap();
+}
+
+/// Generate the SeriesData and SeriesEndpoint classes
+pub fn generate_endpoint_class(output: &mut String) {
+    writeln!(
+        output,
+        r#"# Date conversion constants
+_GENESIS = date(2009, 1, 3)  # day1 0, week1 0
+_DAY_ONE = date(2009, 1, 9)  # day1 1 (6 day gap after genesis)
+_EPOCH = datetime(2009, 1, 1, tzinfo=timezone.utc)
+_DATE_INDEXES = frozenset([
+    'minute10', 'minute30',
+    'hour1', 'hour4', 'hour12',
+    'day1', 'day3', 'week1',
+    'month1', 'month3', 'month6',
+    'year1', 'year10',
+])
+
+def _index_to_date(index: str, i: int) -> Union[date, datetime]:
+    """Convert an index value to a date/datetime for date-based indexes."""
+    if index == 'minute10':
+        return _EPOCH + timedelta(minutes=i * 10)
+    elif index == 'minute30':
+        return _EPOCH + timedelta(minutes=i * 30)
+    elif index == 'hour1':
+        return _EPOCH + timedelta(hours=i)
+    elif index == 'hour4':
+        return _EPOCH + timedelta(hours=i * 4)
+    elif index == 'hour12':
+        return _EPOCH + timedelta(hours=i * 12)
+    elif index == 'day1':
+        return _GENESIS if i == 0 else _DAY_ONE + timedelta(days=i - 1)
+    elif index == 'day3':
+        return _EPOCH.date() - timedelta(days=1) + timedelta(days=i * 3)
+    elif index == 'week1':
+        return _GENESIS + timedelta(weeks=i)
+    elif index == 'month1':
+        return date(2009 + i // 12, i % 12 + 1, 1)
+    elif index == 'month3':
+        m = i * 3
+        return date(2009 + m // 12, m % 12 + 1, 1)
+    elif index == 'month6':
+        m = i * 6
+        return date(2009 + m // 12, m % 12 + 1, 1)
+    elif index == 'year1':
+        return date(2009 + i, 1, 1)
+    elif index == 'year10':
+        return date(2009 + i * 10, 1, 1)
+    else:
+        raise ValueError(f"{{index}} is not a date-based index")
+
+
+def _date_to_index(index: str, d: Union[date, datetime]) -> int:
+    """Convert a date/datetime to an index value for date-based indexes.
+
+    Returns the floor index (latest index whose date is <= the given date).
+    For sub-day indexes (minute*, hour*), a plain date is treated as midnight UTC.
+    """
+    if index in ('minute10', 'minute30', 'hour1', 'hour4', 'hour12'):
+        if isinstance(d, datetime):
+            dt = d if d.tzinfo else d.replace(tzinfo=timezone.utc)
+        else:
+            dt = datetime(d.year, d.month, d.day, tzinfo=timezone.utc)
+        secs = int((dt - _EPOCH).total_seconds())
+        div = {{'minute10': 600, 'minute30': 1800,
+               'hour1': 3600, 'hour4': 14400, 'hour12': 43200}}
+        return secs // div[index]
+    dd = d.date() if isinstance(d, datetime) else d
+    if index == 'day1':
+        if dd < _DAY_ONE:
+            return 0
+        return 1 + (dd - _DAY_ONE).days
+    elif index == 'day3':
+        return (dd - date(2008, 12, 31)).days // 3
+    elif index == 'week1':
+        return (dd - _GENESIS).days // 7
+    elif index == 'month1':
+        return (dd.year - 2009) * 12 + (dd.month - 1)
+    elif index == 'month3':
+        return (dd.year - 2009) * 4 + (dd.month - 1) // 3
+    elif index == 'month6':
+        return (dd.year - 2009) * 2 + (dd.month - 1) // 6
+    elif index == 'year1':
+        return dd.year - 2009
+    elif index == 'year10':
+        return (dd.year - 2009) // 10
+    else:
+        raise ValueError(f"{{index}} is not a date-based index")
+
+
+@dataclass
+class SeriesData(Generic[T]):
+    """Series data with range information. Always int-indexed."""
+    version: int
+    index: Index
+    type: str
+    start: int
+    end: int
+    stamp: str
+    data: List[T]
+
+    @property
+    def is_date_based(self) -> bool:
+        """Whether this series uses a date-based index."""
+        return self.index in _DATE_INDEXES
+
+    def indexes(self) -> List[int]:
+        """Get raw index numbers."""
+        return list(range(self.start, self.end))
+
+    def keys(self) -> List[int]:
+        """Get keys as index numbers."""
+        return self.indexes()
+
+    def items(self) -> List[Tuple[int, T]]:
+        """Get (index, value) pairs."""
+        return list(zip(self.indexes(), self.data))
+
+    def to_dict(self) -> Dict[int, T]:
+        """Return {{index: value}} dict."""
+        return dict(zip(self.indexes(), self.data))
+
+    def __iter__(self) -> Iterator[Tuple[int, T]]:
+        """Iterate over (index, value) pairs."""
+        return iter(zip(self.indexes(), self.data))
+
+    def __len__(self) -> int:
+        return len(self.data)
+
+    def to_polars(self) -> pl.DataFrame:
+        """Convert to Polars DataFrame with 'index' and 'value' columns."""
+        try:
+            import polars as pl  # type: ignore[import-not-found]
+        except ImportError:
+            raise ImportError("polars is required: pip install polars")
+        return pl.DataFrame({{"index": self.indexes(), "value": self.data}})
+
+    def to_pandas(self) -> pd.DataFrame:
+        """Convert to Pandas DataFrame with 'index' and 'value' columns."""
+        try:
+            import pandas as pd  # type: ignore[import-not-found]
+        except ImportError:
+            raise ImportError("pandas is required: pip install pandas")
+        return pd.DataFrame({{"index": self.indexes(), "value": self.data}})
+
+
+@dataclass
+class DateSeriesData(SeriesData[T]):
+    """Series data with date-based index. Extends SeriesData with date methods."""
+
+    def dates(self) -> List[Union[date, datetime]]:
+        """Get dates for the index range. Returns datetime for sub-daily indexes, date for daily+."""
+        return [_index_to_date(self.index, i) for i in range(self.start, self.end)]
+
+    def date_items(self) -> List[Tuple[Union[date, datetime], T]]:
+        """Get (date, value) pairs."""
+        return list(zip(self.dates(), self.data))
+
+    def to_date_dict(self) -> Dict[Union[date, datetime], T]:
+        """Return {{date: value}} dict."""
+        return dict(zip(self.dates(), self.data))
+
+    def to_polars(self, with_dates: bool = True) -> pl.DataFrame:
+        """Convert to Polars DataFrame.
+
+        Returns a DataFrame with columns:
+        - 'date' and 'value' if with_dates=True (default)
+        - 'index' and 'value' otherwise
+        """
+        try:
+            import polars as pl  # type: ignore[import-not-found]
+        except ImportError:
+            raise ImportError("polars is required: pip install polars")
+        if with_dates:
+            return pl.DataFrame({{"date": self.dates(), "value": self.data}})
+        return pl.DataFrame({{"index": self.indexes(), "value": self.data}})
+
+    def to_pandas(self, with_dates: bool = True) -> pd.DataFrame:
+        """Convert to Pandas DataFrame.
+
+        Returns a DataFrame with columns:
+        - 'date' and 'value' if with_dates=True (default)
+        - 'index' and 'value' otherwise
+        """
+        try:
+            import pandas as pd  # type: ignore[import-not-found]
+        except ImportError:
+            raise ImportError("pandas is required: pip install pandas")
+        if with_dates:
+            return pd.DataFrame({{"date": self.dates(), "value": self.data}})
+        return pd.DataFrame({{"index": self.indexes(), "value": self.data}})
+
+
+# Type aliases for non-generic usage
+AnySeriesData = SeriesData[Any]
+AnyDateSeriesData = DateSeriesData[Any]
+
+
+class _EndpointConfig:
+    """Shared endpoint configuration."""
+    client: BrkClient
+    name: str
+    index: Index
+    start: Optional[int]
+    end: Optional[int]
+
+    def __init__(self, client: BrkClient, name: str, index: Index,
+                 start: Optional[int] = None, end: Optional[int] = None):
+        self.client = client
+        self.name = name
+        self.index = index
+        self.start = start
+        self.end = end
+
+    def path(self) -> str:
+        return f"/api/series/{{self.name}}/{{self.index}}"
+
+    def _build_path(self, format: Optional[str] = None) -> str:
+        params = []
+        if self.start is not None:
+            params.append(f"start={{self.start}}")
+        if self.end is not None:
+            params.append(f"end={{self.end}}")
+        if format is not None:
+            params.append(f"format={{format}}")
+        query = "&".join(params)
+        p = self.path()
+        return f"{{p}}?{{query}}" if query else p
+
+    def _new(self, start: Optional[int] = None, end: Optional[int] = None) -> _EndpointConfig:
+        return _EndpointConfig(self.client, self.name, self.index, start, end)
+
+    def get_series(self) -> SeriesData[Any]:
+        return SeriesData(**self.client.get_json(self._build_path()))
+
+    def get_date_series(self) -> DateSeriesData[Any]:
+        return DateSeriesData(**self.client.get_json(self._build_path()))
+
+    def get_csv(self) -> str:
+        return self.client.get_text(self._build_path(format='csv'))
+
+    def get_len(self) -> int:
+        return self.client.get_series_len(self.name, self.index)
+
+    def get_version(self) -> Version:
+        return self.client.get_series_version(self.name, self.index)
+
+
+class RangeBuilder(Generic[T]):
+    """Builder with range specified."""
+
+    def __init__(self, config: _EndpointConfig):
+        self._config = config
+
+    def fetch(self) -> SeriesData[T]:
+        """Fetch the range as parsed JSON."""
+        return self._config.get_series()
+
+    def fetch_csv(self) -> str:
+        """Fetch the range as CSV string."""
+        return self._config.get_csv()
+
+
+class SingleItemBuilder(Generic[T]):
+    """Builder for single item access."""
+
+    def __init__(self, config: _EndpointConfig):
+        self._config = config
+
+    def fetch(self) -> SeriesData[T]:
+        """Fetch the single item."""
+        return self._config.get_series()
+
+    def fetch_csv(self) -> str:
+        """Fetch as CSV."""
+        return self._config.get_csv()
+
+
+class SkippedBuilder(Generic[T]):
+    """Builder after calling skip(n). Chain with take() to specify count."""
+
+    def __init__(self, config: _EndpointConfig):
+        self._config = config
+
+    def take(self, n: int) -> RangeBuilder[T]:
+        """Take n items after the skipped position."""
+        start = self._config.start or 0
+        return RangeBuilder(self._config._new(start, start + n))
+
+    def fetch(self) -> SeriesData[T]:
+        """Fetch from skipped position to end."""
+        return self._config.get_series()
+
+    def fetch_csv(self) -> str:
+        """Fetch as CSV."""
+        return self._config.get_csv()
+
+
+class DateRangeBuilder(RangeBuilder[T]):
+    """Range builder that returns DateSeriesData."""
+    def fetch(self) -> DateSeriesData[T]:
+        return self._config.get_date_series()
+
+
+class DateSingleItemBuilder(SingleItemBuilder[T]):
+    """Single item builder that returns DateSeriesData."""
+    def fetch(self) -> DateSeriesData[T]:
+        return self._config.get_date_series()
+
+
+class DateSkippedBuilder(SkippedBuilder[T]):
+    """Skipped builder that returns DateSeriesData."""
+    def take(self, n: int) -> DateRangeBuilder[T]:
+        start = self._config.start or 0
+        return DateRangeBuilder(self._config._new(start, start + n))
+    def fetch(self) -> DateSeriesData[T]:
+        return self._config.get_date_series()
+
+
+class SeriesEndpoint(Generic[T]):
+    """Builder for series endpoint queries with int-based indexing.
+
+    Examples:
+        data = endpoint.fetch()
+        data = endpoint[5].fetch()
+        data = endpoint[:10].fetch()
+        data = endpoint.head(20).fetch()
+        data = endpoint.skip(100).take(10).fetch()
+    """
+
+    def __init__(self, client: BrkClient, name: str, index: Index):
+        self._config = _EndpointConfig(client, name, index)
+
+    @overload
+    def __getitem__(self, key: int) -> SingleItemBuilder[T]: ...
+    @overload
+    def __getitem__(self, key: slice) -> RangeBuilder[T]: ...
+
+    def __getitem__(self, key: Union[int, slice]) -> Union[SingleItemBuilder[T], RangeBuilder[T]]:
+        """Access single item or slice by integer index."""
+        if isinstance(key, int):
+            return SingleItemBuilder(self._config._new(key, key + 1))
+        return RangeBuilder(self._config._new(key.start, key.stop))
+
+    def head(self, n: int = 10) -> RangeBuilder[T]:
+        """Get the first n items."""
+        return RangeBuilder(self._config._new(end=n))
+
+    def tail(self, n: int = 10) -> RangeBuilder[T]:
+        """Get the last n items."""
+        return RangeBuilder(self._config._new(end=0) if n == 0 else self._config._new(start=-n))
+
+    def skip(self, n: int) -> SkippedBuilder[T]:
+        """Skip the first n items."""
+        return SkippedBuilder(self._config._new(start=n))
+
+    def fetch(self) -> SeriesData[T]:
+        """Fetch all data."""
+        return self._config.get_series()
+
+    def fetch_csv(self) -> str:
+        """Fetch all data as CSV."""
+        return self._config.get_csv()
+
+    def len(self) -> int:
+        """Total number of data points for this series."""
+        return self._config.get_len()
+
+    def version(self) -> Version:
+        """Current version of the series."""
+        return self._config.get_version()
+
+    def path(self) -> str:
+        """Get the base endpoint path."""
+        return self._config.path()
+
+
+class DateSeriesEndpoint(Generic[T]):
+    """Builder for series endpoint queries with date-based indexing.
+
+    Accepts dates in __getitem__ and returns DateSeriesData from fetch().
+
+    Examples:
+        data = endpoint.fetch()
+        data = endpoint[date(2020, 1, 1)].fetch()
+        data = endpoint[date(2020, 1, 1):date(2023, 1, 1)].fetch()
+        data = endpoint[:10].fetch()
+    """
+
+    def __init__(self, client: BrkClient, name: str, index: Index):
+        self._config = _EndpointConfig(client, name, index)
+
+    @overload
+    def __getitem__(self, key: int) -> DateSingleItemBuilder[T]: ...
+    @overload
+    def __getitem__(self, key: datetime) -> DateSingleItemBuilder[T]: ...
+    @overload
+    def __getitem__(self, key: date) -> DateSingleItemBuilder[T]: ...
+    @overload
+    def __getitem__(self, key: slice) -> DateRangeBuilder[T]: ...
+
+    def __getitem__(self, key: Union[int, slice, date, datetime]) -> Union[DateSingleItemBuilder[T], DateRangeBuilder[T]]:
+        """Access single item or slice. Accepts int, date, or datetime."""
+        if isinstance(key, (date, datetime)):
+            idx = _date_to_index(self._config.index, key)
+            return DateSingleItemBuilder(self._config._new(idx, idx + 1))
+        if isinstance(key, int):
+            return DateSingleItemBuilder(self._config._new(key, key + 1))
+        start, stop = key.start, key.stop
+        if isinstance(start, (date, datetime)):
+            start = _date_to_index(self._config.index, start)
+        if isinstance(stop, (date, datetime)):
+            stop = _date_to_index(self._config.index, stop)
+        return DateRangeBuilder(self._config._new(start, stop))
+
+    def head(self, n: int = 10) -> DateRangeBuilder[T]:
+        """Get the first n items."""
+        return DateRangeBuilder(self._config._new(end=n))
+
+    def tail(self, n: int = 10) -> DateRangeBuilder[T]:
+        """Get the last n items."""
+        return DateRangeBuilder(self._config._new(end=0) if n == 0 else self._config._new(start=-n))
+
+    def skip(self, n: int) -> DateSkippedBuilder[T]:
+        """Skip the first n items."""
+        return DateSkippedBuilder(self._config._new(start=n))
+
+    def fetch(self) -> DateSeriesData[T]:
+        """Fetch all data."""
+        return self._config.get_date_series()
+
+    def fetch_csv(self) -> str:
+        """Fetch all data as CSV."""
+        return self._config.get_csv()
+
+    def len(self) -> int:
+        """Total number of data points for this series."""
+        return self._config.get_len()
+
+    def version(self) -> Version:
+        """Current version of the series."""
+        return self._config.get_version()
+
+    def path(self) -> str:
+        """Get the base endpoint path."""
+        return self._config.path()
+
+
+# Type aliases for non-generic usage
+AnySeriesEndpoint = SeriesEndpoint[Any]
+AnyDateSeriesEndpoint = DateSeriesEndpoint[Any]
+
+
+class SeriesPattern(Protocol[T]):
+    """Protocol for series patterns with different index sets."""
+
+    @property
+    def name(self) -> str:
+        """Get the series name."""
+        ...
+
+    def indexes(self) -> List[str]:
+        """Get the list of available indexes for this series."""
+        ...
+
+    def get(self, index: Index) -> Optional[SeriesEndpoint[T]]:
+        """Get an endpoint builder for a specific index, if supported."""
+        ...
+
+"#
+    )
+    .unwrap();
+}
+
+/// Generate index accessor classes
+pub fn generate_index_accessors(output: &mut String, patterns: &[IndexSetPattern]) {
+    if patterns.is_empty() {
+        return;
+    }
+
+    // Generate static index tuples
+    writeln!(output, "# Static index tuples").unwrap();
+    for (i, pattern) in patterns.iter().enumerate() {
+        write!(output, "_i{} = (", i + 1).unwrap();
+        for (j, index) in pattern.indexes.iter().enumerate() {
+            if j > 0 {
+                write!(output, ", ").unwrap();
+            }
+            write!(output, "'{}'", index.name()).unwrap();
+        }
+        // Single-element tuple needs trailing comma
+        if pattern.indexes.len() == 1 {
+            write!(output, ",").unwrap();
+        }
+        writeln!(output, ")").unwrap();
+    }
+    writeln!(output).unwrap();
+
+    // Generate helper functions
+    writeln!(
+        output,
+        r#"def _ep(c: BrkClient, n: str, i: Index) -> SeriesEndpoint[Any]:
+    return SeriesEndpoint(c, n, i)
+
+def _dep(c: BrkClient, n: str, i: Index) -> DateSeriesEndpoint[Any]:
+    return DateSeriesEndpoint(c, n, i)
+"#
+    )
+    .unwrap();
+
+    writeln!(output, "# Index accessor classes\n").unwrap();
+
+    for (i, pattern) in patterns.iter().enumerate() {
+        let by_class_name = format!("_{}By", pattern.name);
+        let idx_var = format!("_i{}", i + 1);
+
+        // Generate the By class with compact methods
+        writeln!(output, "class {}(Generic[T]):", by_class_name).unwrap();
+        writeln!(
+            output,
+            "    def __init__(self, c: BrkClient, n: str): self._c, self._n = c, n"
+        )
+        .unwrap();
+        for index in &pattern.indexes {
+            let method_name = index_to_field_name(index);
+            let index_name = index.name();
+            let (builder_type, helper) = if index.is_date_based() {
+                ("DateSeriesEndpoint", "_dep")
+            } else {
+                ("SeriesEndpoint", "_ep")
+            };
+            writeln!(
+                output,
+                "    def {}(self) -> {}[T]: return {}(self._c, self._n, '{}')",
+                method_name, builder_type, helper, index_name
+            )
+            .unwrap();
+        }
+        writeln!(output).unwrap();
+
+        // Generate the main accessor class
+        writeln!(output, "class {}(Generic[T]):", pattern.name).unwrap();
+        writeln!(output, "    by: {}[T]", by_class_name).unwrap();
+        writeln!(
+            output,
+            "    def __init__(self, c: BrkClient, n: str): self._n, self.by = n, {}(c, n)",
+            by_class_name
+        )
+        .unwrap();
+        writeln!(output, "    @property").unwrap();
+        writeln!(output, "    def name(self) -> str: return self._n").unwrap();
+        writeln!(
+            output,
+            "    def indexes(self) -> List[str]: return list({})",
+            idx_var
+        )
+        .unwrap();
+        writeln!(
+            output,
+            "    def get(self, index: Index) -> Optional[SeriesEndpoint[T]]: return _ep(self.by._c, self._n, index) if index in {} else None",
+            idx_var
+        )
+        .unwrap();
+        writeln!(output).unwrap();
+    }
+}
+
+/// Generate structural pattern classes
+pub fn generate_structural_patterns(
+    output: &mut String,
+    patterns: &[StructuralPattern],
+    metadata: &ClientMetadata,
+) {
+    if patterns.is_empty() {
+        return;
+    }
+
+    writeln!(output, "# Reusable structural pattern classes\n").unwrap();
+
+    for pattern in patterns {
+        // Generate class
+        if pattern.is_generic {
+            writeln!(output, "class {}(Generic[T]):", pattern.name).unwrap();
+        } else {
+            writeln!(output, "class {}:", pattern.name).unwrap();
+        }
+        writeln!(
+            output,
+            "    \"\"\"Pattern struct for repeated tree structure.\"\"\""
+        )
+        .unwrap();
+
+        // Skip constructor for non-parameterizable patterns (inlined at tree level)
+        if !metadata.is_parameterizable(&pattern.name) {
+            writeln!(output, "    pass\n").unwrap();
+            continue;
+        }
+
+        writeln!(output, "    ").unwrap();
+        if pattern.is_templated() {
+            writeln!(
+                output,
+                "    def __init__(self, client: BrkClient, acc: str, disc: str):"
+            )
+            .unwrap();
+        } else {
+            writeln!(
+                output,
+                "    def __init__(self, client: BrkClient, acc: str):"
+            )
+            .unwrap();
+        }
+        writeln!(
+            output,
+            "        \"\"\"Create pattern node with accumulated series name.\"\"\""
+        )
+        .unwrap();
+
+        let syntax = PythonSyntax;
+        for field in &pattern.fields {
+            generate_parameterized_field(output, &syntax, field, pattern, metadata, "        ");
+        }
+
+        writeln!(output).unwrap();
+    }
+}

crates/brk_bindgen/src/generators/python/mod.rs

@@ -0,0 +1,71 @@
+//! Python client generation.
+//!
+//! This module generates a Python client with type hints for the BRK API.
+
+pub mod api;
+pub mod client;
+pub mod tree;
+pub mod types;
+
+use std::{fmt::Write, io, path::Path};
+
+use super::write_if_changed;
+use crate::{ClientMetadata, Endpoint, TypeSchemas};
+
+/// Generate Python client from metadata and OpenAPI endpoints.
+///
+/// `output_path` is the full path to the output file (e.g., "packages/brk_client/__init__.py").
+pub fn generate_python_client(
+    metadata: &ClientMetadata,
+    endpoints: &[Endpoint],
+    schemas: &TypeSchemas,
+    output_path: &Path,
+) -> io::Result<()> {
+    let mut output = String::new();
+
+    writeln!(output, "# Auto-generated BRK Python client").unwrap();
+    writeln!(output, "# Do not edit manually\n").unwrap();
+    writeln!(output, "from __future__ import annotations").unwrap();
+    writeln!(output, "from dataclasses import dataclass").unwrap();
+    writeln!(
+        output,
+        "from typing import TypeVar, Generic, Any, Dict, Optional, List, Iterator, Literal, TypedDict, Union, Protocol, overload, Tuple, TYPE_CHECKING"
+    )
+    .unwrap();
+    writeln!(
+        output,
+        "from http.client import HTTPSConnection, HTTPConnection"
+    )
+    .unwrap();
+    writeln!(output, "from urllib.parse import urlparse").unwrap();
+    writeln!(
+        output,
+        "from datetime import date, datetime, timedelta, timezone"
+    )
+    .unwrap();
+    writeln!(output, "import json\n").unwrap();
+    writeln!(output, "if TYPE_CHECKING:").unwrap();
+    writeln!(
+        output,
+        "    import pandas as pd  # type: ignore[import-not-found]"
+    )
+    .unwrap();
+    writeln!(
+        output,
+        "    import polars as pl  # type: ignore[import-not-found]\n"
+    )
+    .unwrap();
+    writeln!(output, "T = TypeVar('T')\n").unwrap();
+
+    types::generate_type_definitions(&mut output, schemas);
+    client::generate_base_client(&mut output);
+    client::generate_endpoint_class(&mut output);
+    client::generate_index_accessors(&mut output, &metadata.index_set_patterns);
+    client::generate_structural_patterns(&mut output, &metadata.structural_patterns, metadata);
+    tree::generate_tree_classes(&mut output, &metadata.catalog, metadata);
+    api::generate_main_client(&mut output, endpoints);
+
+    write_if_changed(output_path, &output)?;
+
+    Ok(())
+}

crates/brk_bindgen/src/generators/python/tree.rs

@@ -0,0 +1,105 @@
+//! Python tree structure generation.
+
+use std::collections::BTreeSet;
+use std::fmt::Write;
+
+use brk_types::TreeNode;
+
+use crate::{
+    ClientMetadata, LanguageSyntax, PatternField, PythonSyntax, build_child_path,
+    generate_leaf_field, generate_tree_node_field, prepare_tree_node,
+};
+
+/// Generate tree classes
+pub fn generate_tree_classes(output: &mut String, catalog: &TreeNode, metadata: &ClientMetadata) {
+    writeln!(output, "# Series tree classes\n").unwrap();
+
+    let pattern_lookup = metadata.pattern_lookup();
+    let mut generated = BTreeSet::new();
+    generate_tree_class(
+        output,
+        "SeriesTree",
+        "",
+        catalog,
+        pattern_lookup,
+        metadata,
+        &mut generated,
+    );
+}
+
+/// Recursively generate tree classes
+fn generate_tree_class(
+    output: &mut String,
+    name: &str,
+    path: &str,
+    node: &TreeNode,
+    pattern_lookup: &std::collections::BTreeMap<Vec<PatternField>, String>,
+    metadata: &ClientMetadata,
+    generated: &mut BTreeSet<String>,
+) {
+    let Some(ctx) = prepare_tree_node(node, name, path, pattern_lookup, metadata, generated) else {
+        return;
+    };
+
+    // Generate child classes FIRST (post-order traversal)
+    // This ensures children are defined before parent references them
+    for child in &ctx.children {
+        if child.should_inline {
+            let child_path = build_child_path(path, child.name);
+            generate_tree_class(
+                output,
+                &child.inline_type_name,
+                &child_path,
+                child.node,
+                pattern_lookup,
+                metadata,
+                generated,
+            );
+        }
+    }
+
+    // THEN generate the current class (after all children are defined)
+    writeln!(output, "class {}:", name).unwrap();
+    writeln!(output, "    \"\"\"Series tree node.\"\"\"").unwrap();
+    writeln!(output, "    ").unwrap();
+    writeln!(
+        output,
+        "    def __init__(self, client: BrkClient, base_path: str = ''):"
+    )
+    .unwrap();
+
+    if ctx.children.is_empty() {
+        writeln!(output, "        pass").unwrap();
+    }
+
+    let syntax = PythonSyntax;
+    for child in &ctx.children {
+        if child.is_leaf {
+            if let TreeNode::Leaf(leaf) = child.node {
+                generate_leaf_field(
+                    output, &syntax, "client", child.name, leaf, metadata, "        ",
+                );
+            }
+        } else if child.should_inline {
+            let field_name = syntax.field_name(child.name);
+            writeln!(
+                output,
+                "        self.{}: {} = {}(client)",
+                field_name, child.inline_type_name, child.inline_type_name
+            )
+            .unwrap();
+        } else {
+            generate_tree_node_field(
+                output,
+                &syntax,
+                &child.field,
+                metadata,
+                "        ",
+                "client",
+                &child.base_result,
+            );
+        }
+    }
+
+    writeln!(output).unwrap();
+}

crates/brk_bindgen/src/generators/python/types.rs

@@ -0,0 +1,340 @@
+//! Python type definitions generation.
+
+use std::collections::{BTreeMap, BTreeSet};
+use std::fmt::Write;
+
+use serde_json::Value;
+
+use crate::{
+    TypeSchemas, escape_python_keyword, generators::MANUAL_GENERIC_TYPES, get_union_variants,
+    ref_to_type_name,
+};
+
+/// Generate type definitions from schemas.
+pub fn generate_type_definitions(output: &mut String, schemas: &TypeSchemas) {
+    if schemas.is_empty() {
+        return;
+    }
+
+    writeln!(output, "# Type definitions\n").unwrap();
+
+    let sorted_names = topological_sort_schemas(schemas);
+
+    // Partition into simple type aliases and TypedDict classes
+    // Generate type aliases first to avoid forward reference issues
+    let (type_aliases, typed_dicts): (Vec<_>, Vec<_>) = sorted_names
+        .into_iter()
+        .filter(|name| !MANUAL_GENERIC_TYPES.contains(&name.as_str()))
+        .filter(|name| schemas.contains_key(name))
+        .partition(|name| {
+            schemas
+                .get(name)
+                .map(|s| s.get("properties").is_none())
+                .unwrap_or(false)
+        });
+
+    // Generate simple type aliases first
+    // Quote references to TypedDicts since they're defined after
+    let typed_dict_set: BTreeSet<_> = typed_dicts.iter().cloned().collect();
+    for name in type_aliases {
+        let schema = &schemas[&name];
+        let type_desc = schema.get("description").and_then(|d| d.as_str());
+        let py_type = schema_to_python_type(schema, Some(&name), Some(&typed_dict_set));
+        if let Some(desc) = type_desc {
+            for line in desc.lines() {
+                writeln!(output, "# {}", line).unwrap();
+            }
+        }
+        writeln!(output, "{} = {}", name, py_type).unwrap();
+    }
+
+    // Then generate TypedDict classes
+    for name in typed_dicts {
+        let schema = &schemas[&name];
+        let type_desc = schema.get("description").and_then(|d| d.as_str());
+        let props = schema
+            .get("properties")
+            .and_then(|p| p.as_object())
+            .unwrap();
+
+        writeln!(output, "class {}(TypedDict):", name).unwrap();
+
+        // Collect field descriptions for Attributes section
+        let field_docs: Vec<(String, Option<&str>)> = props
+            .iter()
+            .map(|(prop_name, prop_schema)| {
+                let safe_name = escape_python_keyword(prop_name);
+                let desc = prop_schema.get("description").and_then(|d| d.as_str());
+                (safe_name, desc)
+            })
+            .collect();
+        let has_field_docs = field_docs.iter().any(|(_, d)| d.is_some());
+
+        // Generate docstring if we have type description or field descriptions
+        if type_desc.is_some() || has_field_docs {
+            writeln!(output, "    \"\"\"").unwrap();
+            if let Some(desc) = type_desc {
+                for line in desc.lines() {
+                    writeln!(output, "    {}", line).unwrap();
+                }
+            }
+            if has_field_docs {
+                if type_desc.is_some() {
+                    writeln!(output).unwrap();
+                }
+                writeln!(output, "    Attributes:").unwrap();
+                for (field_name, desc) in &field_docs {
+                    if let Some(d) = desc {
+                        writeln!(output, "        {}: {}", field_name, d).unwrap();
+                    }
+                }
+            }
+            writeln!(output, "    \"\"\"").unwrap();
+        }
+
+        for (prop_name, prop_schema) in props {
+            let prop_type = schema_to_python_type(prop_schema, Some(&name), None);
+            let safe_name = escape_python_keyword(prop_name);
+            writeln!(output, "    {}: {}", safe_name, prop_type).unwrap();
+        }
+        writeln!(output).unwrap();
+    }
+    writeln!(output).unwrap();
+}
+
+/// Topologically sort schema names so dependencies come before dependents (avoids forward references).
+/// Types that reference other types (via $ref) must be defined after their dependencies.
+fn topological_sort_schemas(schemas: &TypeSchemas) -> Vec<String> {
+    // Build dependency graph
+    let mut deps: BTreeMap<String, BTreeSet<String>> = BTreeMap::new();
+    for (name, schema) in schemas {
+        let mut type_deps = BTreeSet::new();
+        collect_schema_refs(schema, &mut type_deps);
+        // Only keep deps that are in our schemas, and drop self-references
+        // (handled at emit time by quoting via current_type)
+        type_deps.retain(|d| schemas.contains_key(d) && d != name);
+        deps.insert(name.clone(), type_deps);
+    }
+
+    // Kahn's algorithm for topological sort
+    let mut in_degree: BTreeMap<String, usize> = BTreeMap::new();
+    for name in schemas.keys() {
+        in_degree.insert(name.clone(), 0);
+    }
+    for type_deps in deps.values() {
+        for dep in type_deps {
+            *in_degree.entry(dep.clone()).or_insert(0) += 1;
+        }
+    }
+
+    // Start with types that have no dependents (are not referenced by others)
+    let mut queue: Vec<String> = in_degree
+        .iter()
+        .filter(|(_, count)| **count == 0)
+        .map(|(name, _)| name.clone())
+        .collect();
+    queue.sort(); // Deterministic order
+
+    let mut result = Vec::new();
+    while let Some(name) = queue.pop() {
+        result.push(name.clone());
+        if let Some(type_deps) = deps.get(&name) {
+            for dep in type_deps {
+                if let Some(count) = in_degree.get_mut(dep) {
+                    *count = count.saturating_sub(1);
+                    if *count == 0 {
+                        queue.push(dep.clone());
+                        queue.sort(); // Keep sorted for determinism
+                    }
+                }
+            }
+        }
+    }
+
+    // Reverse so dependencies come first
+    result.reverse();
+
+    // Add any types that weren't processed (e.g., due to circular refs or other edge cases)
+    let result_set: BTreeSet<_> = result.iter().cloned().collect();
+    let mut missing: Vec<_> = schemas
+        .keys()
+        .filter(|k| !result_set.contains(*k))
+        .cloned()
+        .collect();
+    missing.sort();
+    result.extend(missing);
+
+    result
+}
+
+/// Collect all type references ($ref) from a schema
+fn collect_schema_refs(schema: &Value, refs: &mut BTreeSet<String>) {
+    match schema {
+        Value::Object(map) => {
+            if let Some(ref_path) = map.get("$ref").and_then(|r| r.as_str())
+                && let Some(type_name) = ref_to_type_name(ref_path)
+            {
+                refs.insert(type_name.to_string());
+            }
+            for value in map.values() {
+                collect_schema_refs(value, refs);
+            }
+        }
+        Value::Array(arr) => {
+            for item in arr {
+                collect_schema_refs(item, refs);
+            }
+        }
+        _ => {}
+    }
+}
+
+/// Convert a single JSON type string to Python type
+fn json_type_to_python(ty: &str, schema: &Value, current_type: Option<&str>) -> String {
+    match ty {
+        "integer" => "int".to_string(),
+        "number" => "float".to_string(),
+        "boolean" => "bool".to_string(),
+        "string" => "str".to_string(),
+        "null" => "None".to_string(),
+        "array" => {
+            let item_type = schema
+                .get("items")
+                .map(|s| schema_to_python_type(s, current_type, None))
+                .unwrap_or_else(|| "Any".to_string());
+            format!("List[{}]", item_type)
+        }
+        "object" => {
+            if let Some(add_props) = schema.get("additionalProperties") {
+                let value_type = schema_to_python_type(add_props, current_type, None);
+                return format!("dict[str, {}]", value_type);
+            }
+            "dict".to_string()
+        }
+        _ => "Any".to_string(),
+    }
+}
+
+/// Convert JSON Schema to Python type.
+///
+/// - `current_type`: Used to detect and quote self-references for recursive types
+/// - `quote_types`: Optional set of additional type names that should be quoted
+pub fn schema_to_python_type(
+    schema: &Value,
+    current_type: Option<&str>,
+    quote_types: Option<&BTreeSet<String>>,
+) -> String {
+    if let Some(all_of) = schema.get("allOf").and_then(|v| v.as_array()) {
+        for item in all_of {
+            let resolved = schema_to_python_type(item, current_type, quote_types);
+            if resolved != "Any" {
+                return resolved;
+            }
+        }
+    }
+
+    // Handle $ref
+    if let Some(ref_path) = schema.get("$ref").and_then(|r| r.as_str()) {
+        let type_name = ref_to_type_name(ref_path).unwrap_or("Any");
+        // Quote self-references or types in quote_types set
+        let should_quote =
+            current_type == Some(type_name) || quote_types.is_some_and(|qt| qt.contains(type_name));
+        if should_quote {
+            return format!("\"{}\"", type_name);
+        }
+        return type_name.to_string();
+    }
+
+    // Handle enum (array of string values)
+    if let Some(enum_values) = schema.get("enum").and_then(|e| e.as_array()) {
+        let literals: Vec<String> = enum_values
+            .iter()
+            .filter_map(|v| v.as_str())
+            .map(|s| format!("\"{}\"", s))
+            .collect();
+        if !literals.is_empty() {
+            return format!("Literal[{}]", literals.join(", "));
+        }
+    }
+
+    if let Some(ty) = schema.get("type") {
+        if let Some(type_array) = ty.as_array() {
+            let types: Vec<String> = type_array
+                .iter()
+                .filter_map(|t| t.as_str())
+                .filter(|t| *t != "null")
+                .map(|t| json_type_to_python(t, schema, current_type))
+                .collect();
+            let has_null = type_array.iter().any(|t| t.as_str() == Some("null"));
+
+            if types.len() == 1 {
+                let base_type = &types[0];
+                return if has_null {
+                    format!("Optional[{}]", base_type)
+                } else {
+                    base_type.clone()
+                };
+            } else if !types.is_empty() {
+                let union = format!("Union[{}]", types.join(", "));
+                return if has_null {
+                    format!("Optional[{}]", union)
+                } else {
+                    union
+                };
+            }
+        }
+
+        if let Some(ty_str) = ty.as_str() {
+            return json_type_to_python(ty_str, schema, current_type);
+        }
+    }
+
+    if let Some(variants) = get_union_variants(schema) {
+        let types: Vec<String> = variants
+            .iter()
+            .map(|v| schema_to_python_type(v, current_type, quote_types))
+            .collect();
+        let filtered: Vec<_> = types.iter().filter(|t| *t != "Any").collect();
+        if !filtered.is_empty() {
+            return format!(
+                "Union[{}]",
+                filtered
+                    .iter()
+                    .map(|s| s.as_str())
+                    .collect::<Vec<_>>()
+                    .join(", ")
+            );
+        }
+        return format!("Union[{}]", types.join(", "));
+    }
+
+    // Check for format hint without type (common in OpenAPI)
+    if let Some(format) = schema.get("format").and_then(|f| f.as_str()) {
+        return match format {
+            "int32" | "int64" => "int".to_string(),
+            "float" | "double" => "float".to_string(),
+            "date" | "date-time" => "str".to_string(),
+            _ => "Any".to_string(),
+        };
+    }
+
+    "Any".to_string()
+}
+
+/// Convert JS-style type to Python type (e.g., "Txid[]" -> "List[Txid]", "integer" -> "int")
+pub fn js_type_to_python(js_type: &str) -> String {
+    if let Some(inner) = js_type.strip_suffix("[]") {
+        format!("List[{}]", js_type_to_python(inner))
+    } else {
+        match js_type {
+            "integer" => "int".to_string(),
+            "number" => "float".to_string(),
+            "boolean" => "bool".to_string(),
+            "string" => "str".to_string(),
+            "null" => "None".to_string(),
+            "Object" | "object" => "dict".to_string(),
+            "*" => "Any".to_string(),
+            _ => js_type.to_string(),
+        }
+    }
+}

crates/brk_bindgen/src/generators/rust/api.rs

@@ -0,0 +1,335 @@
+//! Rust API method generation.
+
+use std::fmt::Write;
+
+use crate::{Endpoint, VERSION, generators::write_description, to_snake_case};
+
+use super::types::js_type_to_rust;
+
+/// Generate the main BrkClient struct.
+pub fn generate_main_client(output: &mut String, endpoints: &[Endpoint]) {
+    writeln!(
+        output,
+        r#"/// Main BRK client with series tree and API methods.
+pub struct BrkClient {{
+    base: Arc<BrkClientBase>,
+    series: SeriesTree,
+}}
+
+impl BrkClient {{
+    /// Client version.
+    pub const VERSION: &'static str = "v{VERSION}";
+
+    /// Create a new client with the given base URL.
+    pub fn new(base_url: impl Into<String>) -> Self {{
+        let base = Arc::new(BrkClientBase::new(base_url));
+        let series = SeriesTree::new(base.clone(), String::new());
+        Self {{ base, series }}
+    }}
+
+    /// Create a new client with options.
+    pub fn with_options(options: BrkClientOptions) -> Self {{
+        let base = Arc::new(BrkClientBase::with_options(options));
+        let series = SeriesTree::new(base.clone(), String::new());
+        Self {{ base, series }}
+    }}
+
+    /// Get the series tree for navigating series.
+    pub fn series(&self) -> &SeriesTree {{
+        &self.series
+    }}
+
+    /// Create a dynamic series endpoint builder for any series/index combination.
+    ///
+    /// Use this for programmatic access when the series name is determined at runtime.
+    /// For type-safe access, use the `series()` tree instead.
+    ///
+    /// # Example
+    /// ```ignore
+    /// let data = client.series("realized_price", Index::Height)
+    ///     .last(10)
+    ///     .json::<f64>()?;
+    /// ```
+    pub fn series_endpoint(&self, series: impl Into<SeriesName>, index: Index) -> SeriesEndpoint<serde_json::Value> {{
+        SeriesEndpoint::new(
+            self.base.clone(),
+            Arc::from(series.into().as_str()),
+            index,
+        )
+    }}
+
+    /// Create a dynamic date-based series endpoint builder.
+    ///
+    /// Returns `Err` if the index is not date-based.
+    pub fn date_series_endpoint(&self, series: impl Into<SeriesName>, index: Index) -> Result<DateSeriesEndpoint<serde_json::Value>> {{
+        if !index.is_date_based() {{
+            return Err(BrkError {{ message: format!("{{}} is not a date-based index", index.name()) }});
+        }}
+        Ok(DateSeriesEndpoint::new(
+            self.base.clone(),
+            Arc::from(series.into().as_str()),
+            index,
+        ))
+    }}
+"#,
+        VERSION = VERSION
+    )
+    .unwrap();
+
+    generate_api_methods(output, endpoints);
+
+    writeln!(output, "}}").unwrap();
+}
+
+/// Generate API methods from OpenAPI endpoints.
+pub fn generate_api_methods(output: &mut String, endpoints: &[Endpoint]) {
+    for endpoint in endpoints {
+        if !endpoint.should_generate() {
+            continue;
+        }
+        match endpoint.method.as_str() {
+            "GET" => generate_get_method(output, endpoint),
+            "POST" => generate_post_method(output, endpoint),
+            _ => continue,
+        }
+    }
+}
+
+fn generate_get_method(output: &mut String, endpoint: &Endpoint) {
+    let method_name = endpoint_to_method_name(endpoint);
+    let return_type = build_return_type(endpoint);
+
+    write_method_doc(output, endpoint);
+
+    let params = build_method_params(endpoint);
+    writeln!(
+        output,
+        "    pub fn {}(&self{}) -> Result<{}> {{",
+        method_name, params, return_type
+    )
+    .unwrap();
+
+    let (path, index_arg) = build_path_template(endpoint);
+    let fetch_method = if endpoint.returns_binary() {
+        "get_bytes"
+    } else if endpoint.returns_json() {
+        "get_json"
+    } else {
+        "get_text"
+    };
+
+    if endpoint.query_params.is_empty() {
+        writeln!(
+            output,
+            "        self.base.{}(&format!(\"{}\"{}))",
+            fetch_method, path, index_arg
+        )
+        .unwrap();
+    } else {
+        write_query_assembly(output, endpoint, &path, index_arg);
+
+        if endpoint.supports_csv {
+            writeln!(output, "        if format == Some(Format::CSV) {{").unwrap();
+            writeln!(
+                output,
+                "            self.base.get_text(&path).map(FormatResponse::Csv)"
+            )
+            .unwrap();
+            writeln!(output, "        }} else {{").unwrap();
+            writeln!(
+                output,
+                "            self.base.{}(&path).map(FormatResponse::Json)",
+                fetch_method
+            )
+            .unwrap();
+            writeln!(output, "        }}").unwrap();
+        } else {
+            writeln!(output, "        self.base.{}(&path)", fetch_method).unwrap();
+        }
+    }
+
+    writeln!(output, "    }}\n").unwrap();
+}
+
+fn generate_post_method(output: &mut String, endpoint: &Endpoint) {
+    let method_name = endpoint_to_method_name(endpoint);
+    let return_type = build_return_type(endpoint);
+
+    write_method_doc(output, endpoint);
+
+    let mut params = build_method_params(endpoint);
+    if endpoint.request_body.is_some() {
+        params.push_str(", body: &str");
+    }
+    writeln!(
+        output,
+        "    pub fn {}(&self{}) -> Result<{}> {{",
+        method_name, params, return_type
+    )
+    .unwrap();
+
+    let (path, index_arg) = build_path_template(endpoint);
+    let body_arg = if endpoint.request_body.is_some() {
+        "body"
+    } else {
+        "\"\""
+    };
+    let fetch_method = if endpoint.returns_binary() {
+        "post_bytes"
+    } else if endpoint.returns_json() {
+        "post_json"
+    } else {
+        "post_text"
+    };
+
+    if endpoint.query_params.is_empty() {
+        writeln!(
+            output,
+            "        self.base.{}(&format!(\"{}\"{}), {})",
+            fetch_method, path, index_arg, body_arg
+        )
+        .unwrap();
+    } else {
+        write_query_assembly(output, endpoint, &path, index_arg);
+        writeln!(
+            output,
+            "        self.base.{}(&path, {})",
+            fetch_method, body_arg
+        )
+        .unwrap();
+    }
+
+    writeln!(output, "    }}\n").unwrap();
+}
+
+fn build_return_type(endpoint: &Endpoint) -> String {
+    let base = if endpoint.returns_binary() {
+        "Vec<u8>".to_string()
+    } else if endpoint.returns_text() {
+        "String".to_string()
+    } else {
+        endpoint
+            .schema_name()
+            .map(js_type_to_rust)
+            .unwrap_or_else(|| "String".to_string())
+    };
+    if endpoint.supports_csv {
+        format!("FormatResponse<{}>", base)
+    } else {
+        base
+    }
+}
+
+fn write_method_doc(output: &mut String, endpoint: &Endpoint) {
+    let method_name = endpoint_to_method_name(endpoint);
+    writeln!(
+        output,
+        "    /// {}",
+        endpoint.summary.as_deref().unwrap_or(&method_name)
+    )
+    .unwrap();
+    if let Some(desc) = &endpoint.description
+        && endpoint.summary.as_ref() != Some(desc)
+    {
+        writeln!(output, "    ///").unwrap();
+        write_description(output, desc, "    /// ", "    ///");
+    }
+    writeln!(output, "    ///").unwrap();
+    writeln!(
+        output,
+        "    /// Endpoint: `{} {}`",
+        endpoint.method.to_uppercase(),
+        endpoint.path
+    )
+    .unwrap();
+}
+
+fn write_query_assembly(output: &mut String, endpoint: &Endpoint, path: &str, index_arg: &str) {
+    writeln!(output, "        let mut query = Vec::new();").unwrap();
+    for param in &endpoint.query_params {
+        let ident = sanitize_ident(&param.name);
+        let is_array = param.param_type.ends_with("[]");
+        if is_array {
+            writeln!(
+                output,
+                "        for v in {} {{ query.push(format!(\"{}={{}}\", v)); }}",
+                ident, param.name
+            )
+            .unwrap();
+        } else if param.required {
+            writeln!(
+                output,
+                "        query.push(format!(\"{}={{}}\", {}));",
+                param.name, ident
+            )
+            .unwrap();
+        } else {
+            writeln!(
+                output,
+                "        if let Some(v) = {} {{ query.push(format!(\"{}={{}}\", v)); }}",
+                ident, param.name
+            )
+            .unwrap();
+        }
+    }
+    writeln!(output, "        let query_str = if query.is_empty() {{ String::new() }} else {{ format!(\"?{{}}\", query.join(\"&\")) }};").unwrap();
+    writeln!(
+        output,
+        "        let path = format!(\"{}{{}}\"{}, query_str);",
+        path, index_arg
+    )
+    .unwrap();
+}
+
+fn endpoint_to_method_name(endpoint: &Endpoint) -> String {
+    to_snake_case(&endpoint.operation_name())
+}
+
+fn build_method_params(endpoint: &Endpoint) -> String {
+    let mut params = Vec::new();
+    for param in &endpoint.path_params {
+        let rust_type = param_type_to_rust(&param.param_type);
+        params.push(format!(", {}: {}", sanitize_ident(&param.name), rust_type));
+    }
+    for param in &endpoint.query_params {
+        let rust_type = param_type_to_rust(&param.param_type);
+        let name = sanitize_ident(&param.name);
+        if param.required {
+            params.push(format!(", {}: {}", name, rust_type));
+        } else {
+            params.push(format!(", {}: Option<{}>", name, rust_type));
+        }
+    }
+    params.join("")
+}
+
+/// Strip characters invalid in Rust identifiers (e.g. `[]` from `txId[]`).
+fn sanitize_ident(name: &str) -> String {
+    name.replace(['[', ']'], "")
+}
+
+/// Convert parameter type to Rust type for function signatures.
+fn param_type_to_rust(param_type: &str) -> String {
+    if let Some(inner) = param_type.strip_suffix("[]") {
+        return format!("&[{}]", param_type_to_rust(inner));
+    }
+    match param_type {
+        "string" | "*" => "&str".to_string(),
+        "integer" | "number" => "i64".to_string(),
+        "boolean" => "bool".to_string(),
+        other => other.to_string(),
+    }
+}
+
+/// Build path template and extra format args for Index params.
+fn build_path_template(endpoint: &Endpoint) -> (String, &'static str) {
+    let has_index_param = endpoint
+        .path_params
+        .iter()
+        .any(|p| p.name == "index" && p.param_type == "Index");
+    if has_index_param {
+        (endpoint.path.replace("{index}", "{}"), ", index.name()")
+    } else {
+        (endpoint.path.clone(), "")
+    }
+}

crates/brk_bindgen/src/generators/rust/client.rs

@@ -0,0 +1,623 @@
+//! Rust base client and pattern factory generation.
+
+use std::fmt::Write;
+
+use crate::{
+    ClientMetadata, GenericSyntax, IndexSetPattern, RustSyntax, StructuralPattern,
+    escape_rust_keyword, generate_parameterized_field, index_to_field_name, to_snake_case,
+};
+
+/// Generate import statements.
+pub fn generate_imports(output: &mut String) {
+    writeln!(
+        output,
+        r#"use std::sync::Arc;
+use std::ops::{{Bound, RangeBounds}};
+use serde::de::DeserializeOwned;
+pub use brk_cohort::*;
+pub use brk_types::*;
+
+"#
+    )
+    .unwrap();
+}
+
+/// Generate the base BrkClientBase struct and error types.
+pub fn generate_base_client(output: &mut String) {
+    writeln!(
+        output,
+        r#"/// Error type for BRK client operations.
+#[derive(Debug)]
+pub struct BrkError {{
+    pub message: String,
+}}
+
+impl std::fmt::Display for BrkError {{
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {{
+        write!(f, "{{}}", self.message)
+    }}
+}}
+
+impl std::error::Error for BrkError {{}}
+
+/// Result type for BRK client operations.
+pub type Result<T> = std::result::Result<T, BrkError>;
+
+/// Options for configuring the BRK client.
+#[derive(Debug, Clone)]
+pub struct BrkClientOptions {{
+    pub base_url: String,
+    pub timeout_secs: u64,
+}}
+
+impl Default for BrkClientOptions {{
+    fn default() -> Self {{
+        Self {{
+            base_url: "http://localhost:3000".to_string(),
+            timeout_secs: 30,
+        }}
+    }}
+}}
+
+/// Base HTTP client for making requests. Reuses connections via ureq::Agent.
+#[derive(Debug, Clone)]
+pub struct BrkClientBase {{
+    agent: ureq::Agent,
+    base_url: String,
+}}
+
+impl BrkClientBase {{
+    /// Create a new client with the given base URL.
+    pub fn new(base_url: impl Into<String>) -> Self {{
+        Self::with_options(BrkClientOptions {{ base_url: base_url.into(), ..Default::default() }})
+    }}
+
+    /// Create a new client with options.
+    pub fn with_options(options: BrkClientOptions) -> Self {{
+        let agent = ureq::Agent::config_builder()
+            .timeout_global(Some(std::time::Duration::from_secs(options.timeout_secs)))
+            .build()
+            .into();
+        Self {{
+            agent,
+            base_url: options.base_url.trim_end_matches('/').to_string(),
+        }}
+    }}
+
+    fn url(&self, path: &str) -> String {{
+        format!("{{}}{{}}", self.base_url, path)
+    }}
+
+    /// Make a GET request and deserialize JSON response.
+    pub fn get_json<T: DeserializeOwned>(&self, path: &str) -> Result<T> {{
+        self.agent.get(&self.url(path))
+            .call()
+            .and_then(|mut r| r.body_mut().read_json())
+            .map_err(|e| BrkError {{ message: e.to_string() }})
+    }}
+
+    /// Make a GET request and return raw text response.
+    pub fn get_text(&self, path: &str) -> Result<String> {{
+        self.agent.get(&self.url(path))
+            .call()
+            .and_then(|mut r| r.body_mut().read_to_string())
+            .map_err(|e| BrkError {{ message: e.to_string() }})
+    }}
+
+    /// Make a GET request and return raw bytes response.
+    pub fn get_bytes(&self, path: &str) -> Result<Vec<u8>> {{
+        self.agent.get(&self.url(path))
+            .call()
+            .and_then(|mut r| r.body_mut().read_to_vec())
+            .map_err(|e| BrkError {{ message: e.to_string() }})
+    }}
+
+    /// Make a POST request and deserialize JSON response.
+    pub fn post_json<T: DeserializeOwned>(&self, path: &str, body: &str) -> Result<T> {{
+        self.agent.post(&self.url(path))
+            .send(body)
+            .and_then(|mut r| r.body_mut().read_json())
+            .map_err(|e| BrkError {{ message: e.to_string() }})
+    }}
+
+    /// Make a POST request and return raw text response.
+    pub fn post_text(&self, path: &str, body: &str) -> Result<String> {{
+        self.agent.post(&self.url(path))
+            .send(body)
+            .and_then(|mut r| r.body_mut().read_to_string())
+            .map_err(|e| BrkError {{ message: e.to_string() }})
+    }}
+
+    /// Make a POST request and return raw bytes response.
+    pub fn post_bytes(&self, path: &str, body: &str) -> Result<Vec<u8>> {{
+        self.agent.post(&self.url(path))
+            .send(body)
+            .and_then(|mut r| r.body_mut().read_to_vec())
+            .map_err(|e| BrkError {{ message: e.to_string() }})
+    }}
+}}
+
+/// Build series name with suffix.
+#[inline]
+fn _m(acc: &str, s: &str) -> String {{
+    if s.is_empty() {{ acc.to_string() }}
+    else if acc.is_empty() {{ s.to_string() }}
+    else {{ format!("{{acc}}_{{s}}") }}
+}}
+
+/// Build series name with prefix.
+#[inline]
+fn _p(prefix: &str, acc: &str) -> String {{
+    if acc.is_empty() {{ prefix.to_string() }} else {{ format!("{{prefix}}_{{acc}}") }}
+}}
+
+"#
+    )
+    .unwrap();
+}
+
+/// Generate the SeriesPattern trait.
+pub fn generate_series_pattern_trait(output: &mut String) {
+    writeln!(
+        output,
+        r#"/// Non-generic trait for series patterns (usable in collections).
+pub trait AnySeriesPattern {{
+    /// Get the series name.
+    fn name(&self) -> &str;
+
+    /// Get the list of available indexes for this series.
+    fn indexes(&self) -> &'static [Index];
+}}
+
+/// Generic trait for series patterns with endpoint access.
+pub trait SeriesPattern<T>: AnySeriesPattern {{
+    /// Get an endpoint builder for a specific index, if supported.
+    fn get(&self, index: Index) -> Option<SeriesEndpoint<T>>;
+}}
+
+"#
+    )
+    .unwrap();
+}
+
+/// Generate the SeriesEndpoint structs with typestate pattern.
+pub fn generate_endpoint(output: &mut String) {
+    writeln!(
+        output,
+        r#"/// Shared endpoint configuration.
+#[derive(Clone)]
+struct EndpointConfig {{
+    client: Arc<BrkClientBase>,
+    name: Arc<str>,
+    index: Index,
+    start: Option<i64>,
+    end: Option<i64>,
+}}
+
+impl EndpointConfig {{
+    fn new(client: Arc<BrkClientBase>, name: Arc<str>, index: Index) -> Self {{
+        Self {{ client, name, index, start: None, end: None }}
+    }}
+
+    fn path(&self) -> String {{
+        format!("/api/series/{{}}/{{}}", self.name, self.index.name())
+    }}
+
+    fn build_path(&self, format: Option<&str>) -> String {{
+        let mut params = Vec::new();
+        if let Some(s) = self.start {{ params.push(format!("start={{}}", s)); }}
+        if let Some(e) = self.end {{ params.push(format!("end={{}}", e)); }}
+        if let Some(fmt) = format {{ params.push(format!("format={{}}", fmt)); }}
+        let p = self.path();
+        if params.is_empty() {{ p }} else {{ format!("{{}}?{{}}", p, params.join("&")) }}
+    }}
+
+    fn get_json<T: DeserializeOwned>(&self, format: Option<&str>) -> Result<T> {{
+        self.client.get_json(&self.build_path(format))
+    }}
+
+    fn get_text(&self, format: Option<&str>) -> Result<String> {{
+        self.client.get_text(&self.build_path(format))
+    }}
+
+    fn get_len(&self) -> Result<i64> {{
+        self.client.get_json(&format!("/api/series/{{}}/{{}}/len", self.name, self.index.name()))
+    }}
+
+    fn get_version(&self) -> Result<Version> {{
+        self.client.get_json(&format!("/api/series/{{}}/{{}}/version", self.name, self.index.name()))
+    }}
+}}
+
+/// Builder for series endpoint queries.
+///
+/// Parameterized by element type `T` and response type `D` (defaults to `SeriesData<T>`).
+/// For date-based indexes, use `DateSeriesEndpoint<T>` which sets `D = DateSeriesData<T>`.
+///
+/// # Examples
+/// ```ignore
+/// let data = endpoint.fetch()?;                   // all data
+/// let data = endpoint.get(5).fetch()?;             // single item
+/// let data = endpoint.range(..10).fetch()?;        // first 10
+/// let data = endpoint.range(100..200).fetch()?;    // range [100, 200)
+/// let data = endpoint.take(10).fetch()?;           // first 10 (convenience)
+/// let data = endpoint.last(10).fetch()?;           // last 10
+/// let data = endpoint.skip(100).take(10).fetch()?; // iterator-style
+/// ```
+pub struct SeriesEndpoint<T, D = SeriesData<T>> {{
+    config: EndpointConfig,
+    _marker: std::marker::PhantomData<fn() -> (T, D)>,
+}}
+
+/// Builder for date-based series endpoint queries.
+///
+/// Like `SeriesEndpoint` but returns `DateSeriesData` and provides
+/// date-based access methods (`get_date`, `date_range`).
+pub type DateSeriesEndpoint<T> = SeriesEndpoint<T, DateSeriesData<T>>;
+
+impl<T: DeserializeOwned, D: DeserializeOwned> SeriesEndpoint<T, D> {{
+    pub fn new(client: Arc<BrkClientBase>, name: Arc<str>, index: Index) -> Self {{
+        Self {{ config: EndpointConfig::new(client, name, index), _marker: std::marker::PhantomData }}
+    }}
+
+    /// Select a specific index position.
+    pub fn get(mut self, index: usize) -> SingleItemBuilder<T, D> {{
+        self.config.start = Some(index as i64);
+        self.config.end = Some(index as i64 + 1);
+        SingleItemBuilder {{ config: self.config, _marker: std::marker::PhantomData }}
+    }}
+
+    /// Select a range using Rust range syntax.
+    ///
+    /// # Examples
+    /// ```ignore
+    /// endpoint.range(..10)      // first 10
+    /// endpoint.range(100..110)  // indices 100-109
+    /// endpoint.range(100..)     // from 100 to end
+    /// ```
+    pub fn range<R: RangeBounds<usize>>(mut self, range: R) -> RangeBuilder<T, D> {{
+        self.config.start = match range.start_bound() {{
+            Bound::Included(&n) => Some(n as i64),
+            Bound::Excluded(&n) => Some(n as i64 + 1),
+            Bound::Unbounded => None,
+        }};
+        self.config.end = match range.end_bound() {{
+            Bound::Included(&n) => Some(n as i64 + 1),
+            Bound::Excluded(&n) => Some(n as i64),
+            Bound::Unbounded => None,
+        }};
+        RangeBuilder {{ config: self.config, _marker: std::marker::PhantomData }}
+    }}
+
+    /// Take the first n items.
+    pub fn take(self, n: usize) -> RangeBuilder<T, D> {{
+        self.range(..n)
+    }}
+
+    /// Take the last n items.
+    pub fn last(mut self, n: usize) -> RangeBuilder<T, D> {{
+        if n == 0 {{
+            self.config.end = Some(0);
+        }} else {{
+            self.config.start = Some(-(n as i64));
+        }}
+        RangeBuilder {{ config: self.config, _marker: std::marker::PhantomData }}
+    }}
+
+    /// Skip the first n items. Chain with `take(n)` to get a range.
+    pub fn skip(mut self, n: usize) -> SkippedBuilder<T, D> {{
+        self.config.start = Some(n as i64);
+        SkippedBuilder {{ config: self.config, _marker: std::marker::PhantomData }}
+    }}
+
+    /// Fetch all data as parsed JSON.
+    pub fn fetch(self) -> Result<D> {{
+        self.config.get_json(None)
+    }}
+
+    /// Fetch all data as CSV string.
+    pub fn fetch_csv(self) -> Result<String> {{
+        self.config.get_text(Some("csv"))
+    }}
+
+    /// Total number of data points for this series.
+    #[allow(clippy::len_without_is_empty)]
+    pub fn len(&self) -> Result<i64> {{
+        self.config.get_len()
+    }}
+
+    /// Current version of the series.
+    pub fn version(&self) -> Result<Version> {{
+        self.config.get_version()
+    }}
+
+    /// Get the base endpoint path.
+    pub fn path(&self) -> String {{
+        self.config.path()
+    }}
+}}
+
+/// Date-specific methods available only on `DateSeriesEndpoint`.
+impl<T: DeserializeOwned> SeriesEndpoint<T, DateSeriesData<T>> {{
+    /// Select a specific date position (for day-precision or coarser indexes).
+    pub fn get_date(self, date: Date) -> SingleItemBuilder<T, DateSeriesData<T>> {{
+        let index = self.config.index.date_to_index(date).unwrap_or(0);
+        self.get(index)
+    }}
+
+    /// Select a date range (for day-precision or coarser indexes).
+    pub fn date_range(self, start: Date, end: Date) -> RangeBuilder<T, DateSeriesData<T>> {{
+        let s = self.config.index.date_to_index(start).unwrap_or(0);
+        let e = self.config.index.date_to_index(end).unwrap_or(0);
+        self.range(s..e)
+    }}
+
+    /// Select a specific timestamp position (works for all date-based indexes including sub-daily).
+    pub fn get_timestamp(self, ts: Timestamp) -> SingleItemBuilder<T, DateSeriesData<T>> {{
+        let index = self.config.index.timestamp_to_index(ts).unwrap_or(0);
+        self.get(index)
+    }}
+
+    /// Select a timestamp range (works for all date-based indexes including sub-daily).
+    pub fn timestamp_range(self, start: Timestamp, end: Timestamp) -> RangeBuilder<T, DateSeriesData<T>> {{
+        let s = self.config.index.timestamp_to_index(start).unwrap_or(0);
+        let e = self.config.index.timestamp_to_index(end).unwrap_or(0);
+        self.range(s..e)
+    }}
+}}
+
+/// Builder for single item access.
+pub struct SingleItemBuilder<T, D = SeriesData<T>> {{
+    config: EndpointConfig,
+    _marker: std::marker::PhantomData<fn() -> (T, D)>,
+}}
+
+/// Date-aware single item builder.
+pub type DateSingleItemBuilder<T> = SingleItemBuilder<T, DateSeriesData<T>>;
+
+impl<T: DeserializeOwned, D: DeserializeOwned> SingleItemBuilder<T, D> {{
+    /// Fetch the single item.
+    pub fn fetch(self) -> Result<D> {{
+        self.config.get_json(None)
+    }}
+
+    /// Fetch the single item as CSV.
+    pub fn fetch_csv(self) -> Result<String> {{
+        self.config.get_text(Some("csv"))
+    }}
+}}
+
+/// Builder after calling `skip(n)`. Chain with `take(n)` to specify count.
+pub struct SkippedBuilder<T, D = SeriesData<T>> {{
+    config: EndpointConfig,
+    _marker: std::marker::PhantomData<fn() -> (T, D)>,
+}}
+
+/// Date-aware skipped builder.
+pub type DateSkippedBuilder<T> = SkippedBuilder<T, DateSeriesData<T>>;
+
+impl<T: DeserializeOwned, D: DeserializeOwned> SkippedBuilder<T, D> {{
+    /// Take n items after the skipped position.
+    pub fn take(mut self, n: usize) -> RangeBuilder<T, D> {{
+        let start = self.config.start.unwrap_or(0);
+        self.config.end = Some(start + n as i64);
+        RangeBuilder {{ config: self.config, _marker: std::marker::PhantomData }}
+    }}
+
+    /// Fetch from the skipped position to the end.
+    pub fn fetch(self) -> Result<D> {{
+        self.config.get_json(None)
+    }}
+
+    /// Fetch from the skipped position to the end as CSV.
+    pub fn fetch_csv(self) -> Result<String> {{
+        self.config.get_text(Some("csv"))
+    }}
+}}
+
+/// Builder with range fully specified.
+pub struct RangeBuilder<T, D = SeriesData<T>> {{
+    config: EndpointConfig,
+    _marker: std::marker::PhantomData<fn() -> (T, D)>,
+}}
+
+/// Date-aware range builder.
+pub type DateRangeBuilder<T> = RangeBuilder<T, DateSeriesData<T>>;
+
+impl<T: DeserializeOwned, D: DeserializeOwned> RangeBuilder<T, D> {{
+    /// Fetch the range as parsed JSON.
+    pub fn fetch(self) -> Result<D> {{
+        self.config.get_json(None)
+    }}
+
+    /// Fetch the range as CSV string.
+    pub fn fetch_csv(self) -> Result<String> {{
+        self.config.get_text(Some("csv"))
+    }}
+}}
+
+"#
+    )
+    .unwrap();
+}
+
+/// Generate index accessor structs.
+pub fn generate_index_accessors(output: &mut String, patterns: &[IndexSetPattern]) {
+    if patterns.is_empty() {
+        return;
+    }
+
+    // Generate static index arrays
+    writeln!(output, "// Static index arrays").unwrap();
+    for (i, pattern) in patterns.iter().enumerate() {
+        write!(output, "const _I{}: &[Index] = &[", i + 1).unwrap();
+        for (j, index) in pattern.indexes.iter().enumerate() {
+            if j > 0 {
+                write!(output, ", ").unwrap();
+            }
+            write!(output, "Index::{}", index).unwrap();
+        }
+        writeln!(output, "];").unwrap();
+    }
+    writeln!(output).unwrap();
+
+    // Generate helper functions
+    writeln!(
+        output,
+        r#"#[inline]
+fn _ep<T: DeserializeOwned>(c: &Arc<BrkClientBase>, n: &Arc<str>, i: Index) -> SeriesEndpoint<T> {{
+    SeriesEndpoint::new(c.clone(), n.clone(), i)
+}}
+
+#[inline]
+fn _dep<T: DeserializeOwned>(c: &Arc<BrkClientBase>, n: &Arc<str>, i: Index) -> DateSeriesEndpoint<T> {{
+    DateSeriesEndpoint::new(c.clone(), n.clone(), i)
+}}
+"#
+    )
+    .unwrap();
+
+    // Generate index accessor structs
+    writeln!(output, "// Index accessor structs\n").unwrap();
+
+    for (i, pattern) in patterns.iter().enumerate() {
+        let by_name = format!("{}By", pattern.name);
+        let idx_const = format!("_I{}", i + 1);
+
+        // Generate the "By" struct
+        writeln!(output, "pub struct {}<T> {{ client: Arc<BrkClientBase>, name: Arc<str>, _marker: std::marker::PhantomData<T> }}", by_name).unwrap();
+        writeln!(output, "impl<T: DeserializeOwned> {}<T> {{", by_name).unwrap();
+        for index in &pattern.indexes {
+            let method_name = index_to_field_name(index);
+            if index.is_date_based() {
+                writeln!(
+                    output,
+                    "    pub fn {}(&self) -> DateSeriesEndpoint<T> {{ _dep(&self.client, &self.name, Index::{}) }}",
+                    method_name, index
+                )
+                .unwrap();
+            } else {
+                writeln!(
+                    output,
+                    "    pub fn {}(&self) -> SeriesEndpoint<T> {{ _ep(&self.client, &self.name, Index::{}) }}",
+                    method_name, index
+                )
+                .unwrap();
+            }
+        }
+        writeln!(output, "}}\n").unwrap();
+
+        // Generate the main accessor struct
+        writeln!(
+            output,
+            "pub struct {}<T> {{ name: Arc<str>, pub by: {}<T> }}",
+            pattern.name, by_name
+        )
+        .unwrap();
+        writeln!(output, "impl<T: DeserializeOwned> {}<T> {{", pattern.name).unwrap();
+        writeln!(
+            output,
+            "    pub fn new(client: Arc<BrkClientBase>, name: String) -> Self {{ let name: Arc<str> = name.into(); Self {{ name: name.clone(), by: {} {{ client, name, _marker: std::marker::PhantomData }} }} }}",
+            by_name
+        )
+        .unwrap();
+        writeln!(output, "    pub fn name(&self) -> &str {{ &self.name }}").unwrap();
+        writeln!(output, "}}\n").unwrap();
+
+        // Implement AnySeriesPattern trait
+        writeln!(
+            output,
+            "impl<T> AnySeriesPattern for {}<T> {{ fn name(&self) -> &str {{ &self.name }} fn indexes(&self) -> &'static [Index] {{ {} }} }}",
+            pattern.name, idx_const
+        )
+        .unwrap();
+
+        // Implement SeriesPattern<T> trait
+        writeln!(
+            output,
+            "impl<T: DeserializeOwned> SeriesPattern<T> for {}<T> {{ fn get(&self, index: Index) -> Option<SeriesEndpoint<T>> {{ {}.contains(&index).then(|| _ep(&self.by.client, &self.by.name, index)) }} }}\n",
+            pattern.name, idx_const
+        )
+        .unwrap();
+    }
+}
+
+/// Generate structural pattern structs.
+pub fn generate_pattern_structs(
+    output: &mut String,
+    patterns: &[StructuralPattern],
+    metadata: &ClientMetadata,
+) {
+    if patterns.is_empty() {
+        return;
+    }
+
+    writeln!(output, "// Reusable pattern structs\n").unwrap();
+
+    for pattern in patterns {
+        let generic_params = if pattern.is_generic { "<T>" } else { "" };
+
+        // Generate struct definition
+        writeln!(output, "/// Pattern struct for repeated tree structure.").unwrap();
+        writeln!(output, "pub struct {}{} {{", pattern.name, generic_params).unwrap();
+
+        for field in &pattern.fields {
+            let field_name = escape_rust_keyword(&to_snake_case(&field.name));
+            let type_annotation = metadata.field_type_annotation(
+                field,
+                pattern.is_generic,
+                None,
+                GenericSyntax::RUST,
+            );
+            writeln!(output, "    pub {}: {},", field_name, type_annotation).unwrap();
+        }
+
+        writeln!(output, "}}\n").unwrap();
+
+        // Skip constructor for non-parameterizable patterns (inlined at tree level)
+        if !metadata.is_parameterizable(&pattern.name) {
+            continue;
+        }
+
+        let impl_generic = if pattern.is_generic {
+            "<T: DeserializeOwned>"
+        } else {
+            ""
+        };
+        writeln!(
+            output,
+            "impl{} {}{} {{",
+            impl_generic, pattern.name, generic_params
+        )
+        .unwrap();
+
+        writeln!(
+            output,
+            "    /// Create a new pattern node with accumulated series name."
+        )
+        .unwrap();
+        if pattern.is_templated() {
+            writeln!(
+                output,
+                "    pub fn new(client: Arc<BrkClientBase>, acc: String, disc: String) -> Self {{"
+            )
+            .unwrap();
+        } else {
+            writeln!(
+                output,
+                "    pub fn new(client: Arc<BrkClientBase>, acc: String) -> Self {{"
+            )
+            .unwrap();
+        }
+        writeln!(output, "        Self {{").unwrap();
+
+        let syntax = RustSyntax;
+        for field in &pattern.fields {
+            generate_parameterized_field(output, &syntax, field, pattern, metadata, "            ");
+        }
+
+        writeln!(output, "        }}").unwrap();
+        writeln!(output, "    }}").unwrap();
+        writeln!(output, "}}\n").unwrap();
+    }
+}

crates/brk_bindgen/src/generators/rust/mod.rs

@@ -0,0 +1,46 @@
+//! Rust client generation.
+//!
+//! This module generates a Rust client with full type safety for the BRK API.
+
+pub mod api;
+pub mod client;
+pub mod tree;
+mod types;
+
+use std::{fmt::Write, io, path::Path};
+
+use super::write_if_changed;
+use crate::{ClientMetadata, Endpoint};
+
+/// Generate Rust client from metadata and OpenAPI endpoints.
+///
+/// `output_path` is the full path to the output file (e.g., "crates/brk_client/src/lib.rs").
+pub fn generate_rust_client(
+    metadata: &ClientMetadata,
+    endpoints: &[Endpoint],
+    output_path: &Path,
+) -> io::Result<()> {
+    let mut output = String::new();
+
+    writeln!(output, "// Auto-generated BRK Rust client").unwrap();
+    writeln!(output, "// Do not edit manually\n").unwrap();
+    writeln!(output, "#![allow(non_camel_case_types)]").unwrap();
+    writeln!(output, "#![allow(non_snake_case)]").unwrap();
+    writeln!(output, "#![allow(dead_code)]").unwrap();
+    writeln!(output, "#![allow(unused_variables)]").unwrap();
+    writeln!(output, "#![allow(clippy::useless_format)]").unwrap();
+    writeln!(output, "#![allow(clippy::unnecessary_to_owned)]\n").unwrap();
+
+    client::generate_imports(&mut output);
+    client::generate_base_client(&mut output);
+    client::generate_series_pattern_trait(&mut output);
+    client::generate_endpoint(&mut output);
+    client::generate_index_accessors(&mut output, &metadata.index_set_patterns);
+    client::generate_pattern_structs(&mut output, &metadata.structural_patterns, metadata);
+    tree::generate_tree(&mut output, &metadata.catalog, metadata);
+    api::generate_main_client(&mut output, endpoints);
+
+    write_if_changed(output_path, &output)?;
+
+    Ok(())
+}

crates/brk_bindgen/src/generators/rust/tree.rs

@@ -0,0 +1,126 @@
+//! Rust tree structure generation.
+
+use std::collections::BTreeSet;
+use std::fmt::Write;
+
+use brk_types::TreeNode;
+
+use crate::{
+    ClientMetadata, GenericSyntax, LanguageSyntax, PatternField, RustSyntax, build_child_path,
+    escape_rust_keyword, generate_leaf_field, generate_tree_node_field, prepare_tree_node,
+    to_snake_case,
+};
+
+/// Generate tree structs.
+pub fn generate_tree(output: &mut String, catalog: &TreeNode, metadata: &ClientMetadata) {
+    writeln!(output, "// Series tree\n").unwrap();
+
+    let pattern_lookup = metadata.pattern_lookup();
+    let mut generated = BTreeSet::new();
+    generate_tree_node(
+        output,
+        "SeriesTree",
+        "",
+        catalog,
+        pattern_lookup,
+        metadata,
+        &mut generated,
+    );
+}
+
+fn generate_tree_node(
+    output: &mut String,
+    name: &str,
+    path: &str,
+    node: &TreeNode,
+    pattern_lookup: &std::collections::BTreeMap<Vec<PatternField>, String>,
+    metadata: &ClientMetadata,
+    generated: &mut BTreeSet<String>,
+) {
+    let Some(ctx) = prepare_tree_node(node, name, path, pattern_lookup, metadata, generated) else {
+        return;
+    };
+
+    // Generate struct definition
+    writeln!(output, "/// Series tree node.").unwrap();
+    writeln!(output, "pub struct {} {{", name).unwrap();
+
+    for child in &ctx.children {
+        let field_name = escape_rust_keyword(&to_snake_case(child.name));
+        let type_annotation = if child.should_inline {
+            child.inline_type_name.clone()
+        } else {
+            metadata.field_type_annotation(&child.field, false, None, GenericSyntax::RUST)
+        };
+        writeln!(output, "    pub {}: {},", field_name, type_annotation).unwrap();
+    }
+
+    writeln!(output, "}}\n").unwrap();
+
+    // Generate impl block
+    writeln!(output, "impl {} {{", name).unwrap();
+    writeln!(
+        output,
+        "    pub fn new(client: Arc<BrkClientBase>, base_path: String) -> Self {{"
+    )
+    .unwrap();
+    writeln!(output, "        Self {{").unwrap();
+
+    let syntax = RustSyntax;
+    for child in &ctx.children {
+        let field_name = escape_rust_keyword(&to_snake_case(child.name));
+
+        if child.is_leaf {
+            if let TreeNode::Leaf(leaf) = child.node {
+                generate_leaf_field(
+                    output,
+                    &syntax,
+                    "client.clone()",
+                    child.name,
+                    leaf,
+                    metadata,
+                    "            ",
+                );
+            }
+        } else if child.should_inline {
+            // Inline struct type - only for nodes that don't match any pattern
+            let path_expr = syntax.path_expr("base_path", &format!("_{}", child.name));
+            writeln!(
+                output,
+                "            {}: {}::new(client.clone(), {}),",
+                field_name, child.inline_type_name, path_expr
+            )
+            .unwrap();
+        } else {
+            generate_tree_node_field(
+                output,
+                &syntax,
+                &child.field,
+                metadata,
+                "            ",
+                "client.clone()",
+                &child.base_result,
+            );
+        }
+    }
+
+    writeln!(output, "        }}").unwrap();
+    writeln!(output, "    }}").unwrap();
+    writeln!(output, "}}\n").unwrap();
+
+    // Generate child structs
+    for child in &ctx.children {
+        if child.should_inline {
+            let child_path = build_child_path(path, child.name);
+            generate_tree_node(
+                output,
+                &child.inline_type_name,
+                &child_path,
+                child.node,
+                pattern_lookup,
+                metadata,
+                generated,
+            );
+        }
+    }
+}

crates/brk_bindgen/src/generators/rust/types.rs

@@ -0,0 +1,17 @@
+//! Rust type conversion utilities.
+
+/// Convert JS-style type to Rust type.
+pub fn js_type_to_rust(js_type: &str) -> String {
+    if let Some(inner) = js_type.strip_suffix("[]") {
+        format!("Vec<{}>", js_type_to_rust(inner))
+    } else {
+        match js_type {
+            "string" => "String".to_string(),
+            "integer" => "i64".to_string(),
+            "number" => "f64".to_string(),
+            "boolean" => "bool".to_string(),
+            "*" | "Object" => "serde_json::Value".to_string(),
+            other => other.to_string(),
+        }
+    }
+}

crates/brk_bindgen/src/lib.rs

@@ -0,0 +1,187 @@
+#![allow(clippy::type_complexity)]
+
+use std::{collections::btree_map::Entry, fs::create_dir_all, io, path::PathBuf};
+
+use brk_query::Vecs;
+
+/// Output path configuration for each language client.
+///
+/// Each path should be the full path to the output file, not just a directory.
+/// Parent directories will be created automatically if they don't exist.
+///
+/// # Example
+/// ```ignore
+/// let paths = ClientOutputPaths::new()
+///     .rust("crates/brk_client/src/lib.rs")
+///     .javascript("modules/brk-client/index.js")
+///     .python("packages/brk_client/__init__.py");
+/// ```
+#[derive(Debug, Clone, Default)]
+pub struct ClientOutputPaths {
+    /// Full path to Rust client file (e.g., "crates/brk_client/src/lib.rs")
+    pub rust: Option<PathBuf>,
+    /// Full path to JavaScript client file (e.g., "modules/brk-client/index.js")
+    pub javascript: Option<PathBuf>,
+    /// Full path to Python client file (e.g., "packages/brk_client/__init__.py")
+    pub python: Option<PathBuf>,
+}
+
+impl ClientOutputPaths {
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    pub fn rust(mut self, path: impl Into<PathBuf>) -> Self {
+        self.rust = Some(path.into());
+        self
+    }
+
+    pub fn javascript(mut self, path: impl Into<PathBuf>) -> Self {
+        self.javascript = Some(path.into());
+        self
+    }
+
+    pub fn python(mut self, path: impl Into<PathBuf>) -> Self {
+        self.python = Some(path.into());
+        self
+    }
+}
+
+mod analysis;
+mod backends;
+mod generate;
+mod generators;
+mod openapi;
+mod syntax;
+mod types;
+
+pub use analysis::*;
+pub use backends::*;
+pub use generate::*;
+pub use generators::*;
+pub use openapi::*;
+pub use syntax::*;
+pub use types::*;
+
+pub const VERSION: &str = env!("CARGO_PKG_VERSION");
+
+/// Generate all client libraries from the query vecs and OpenAPI JSON.
+///
+/// Uses `ClientOutputPaths` to specify the output file path for each language.
+/// Only languages with a configured path will be generated.
+///
+/// # Example
+/// ```ignore
+/// let paths = ClientOutputPaths::new()
+///     .rust("crates/brk_client/src/lib.rs")
+///     .javascript("modules/brk-client/index.js")
+///     .python("packages/brk_client/__init__.py");
+///
+/// generate_clients(&vecs, &openapi_json, &paths)?;
+/// ```
+pub fn generate_clients(
+    vecs: &Vecs,
+    openapi_json: &str,
+    output_paths: &ClientOutputPaths,
+) -> io::Result<()> {
+    let metadata = ClientMetadata::from_vecs(vecs);
+
+    // Parse OpenAPI spec
+    let spec = parse_openapi_json(openapi_json)?;
+    let endpoints = extract_endpoints(&spec);
+    let mut schemas = extract_schemas(openapi_json);
+
+    // Collect leaf type schemas from the catalog and merge into schemas
+    collect_leaf_type_schemas(&metadata.catalog, &mut schemas);
+
+    // Also collect definitions from all schemas (including OpenAPI schemas)
+    // We need to do this after collecting leaf schemas so we process everything
+    let schema_values: Vec<_> = schemas.values().cloned().collect();
+    for schema in &schema_values {
+        collect_schema_definitions(schema, &mut schemas);
+    }
+
+    // Generate Rust client (uses real brk_types, no schema conversion needed)
+    if let Some(rust_path) = &output_paths.rust {
+        if let Some(parent) = rust_path.parent() {
+            create_dir_all(parent)?;
+        }
+        generate_rust_client(&metadata, &endpoints, rust_path)?;
+    }
+
+    // Generate JavaScript client (needs schemas for type definitions)
+    if let Some(js_path) = &output_paths.javascript {
+        if let Some(parent) = js_path.parent() {
+            create_dir_all(parent)?;
+        }
+        generate_javascript_client(&metadata, &endpoints, &schemas, js_path)?;
+    }
+
+    // Generate Python client (needs schemas for type definitions)
+    if let Some(python_path) = &output_paths.python {
+        if let Some(parent) = python_path.parent() {
+            create_dir_all(parent)?;
+        }
+        generate_python_client(&metadata, &endpoints, &schemas, python_path)?;
+    }
+
+    Ok(())
+}
+
+use brk_types::TreeNode;
+use serde_json::Value;
+
+/// Recursively collect leaf type schemas from the tree and add to schemas map.
+/// Only adds schemas that aren't already present (OpenAPI schemas take precedence).
+/// Collects definitions from schemars-generated schemas (for referenced types).
+fn collect_leaf_type_schemas(node: &TreeNode, schemas: &mut TypeSchemas) {
+    match node {
+        TreeNode::Leaf(leaf) => {
+            // Collect definitions from the schema (schemars puts type schemas here)
+            // This includes the inner types like `Bitcoin` from `Close<Bitcoin>`
+            collect_schema_definitions(&leaf.schema, schemas);
+
+            // Get the type name for this leaf
+            let type_name = extract_inner_type(leaf.kind());
+
+            if let Entry::Vacant(e) = schemas.entry(type_name) {
+                // Unwrap single-element allOf
+                let schema = unwrap_allof(&leaf.schema);
+
+                // Add the schema if it's usable:
+                // - Simple type (has "type")
+                // - Object type with properties (complex types like OHLCCents, EmptyAddressData)
+                // - Enum type (has "enum" or "oneOf")
+                // - Or a $ref to another type
+                let has_type = schema.get("type").is_some();
+                let has_properties = schema.get("properties").is_some();
+                let has_enum = schema.get("enum").is_some() || schema.get("oneOf").is_some();
+                let is_ref = schema.get("$ref").is_some();
+
+                if has_type || has_properties || has_enum || is_ref {
+                    e.insert(schema.clone());
+                }
+            }
+        }
+        TreeNode::Branch(children) => {
+            for child in children.values() {
+                collect_leaf_type_schemas(child, schemas);
+            }
+        }
+    }
+}
+
+/// Collect type definitions from schemars-generated schema's definitions section.
+/// Schemars uses `definitions` or `$defs` to store referenced types.
+fn collect_schema_definitions(schema: &Value, schemas: &mut TypeSchemas) {
+    // Check both JSON Schema draft-07 style ("definitions") and draft 2019-09+ style ("$defs")
+    for key in ["definitions", "$defs"] {
+        if let Some(defs) = schema.get(key).and_then(|d| d.as_object()) {
+            for (name, def_schema) in defs {
+                if !schemas.contains_key(name) {
+                    schemas.insert(name.clone(), def_schema.clone());
+                }
+            }
+        }
+    }
+}

crates/brk_bindgen/src/openapi/endpoint.rs

@@ -0,0 +1,92 @@
+use crate::openapi::{Parameter, ResponseKind};
+
+/// Request body shape for POST/PUT/PATCH endpoints.
+#[derive(Debug, Clone)]
+pub struct RequestBody {
+    /// Body content type as a name (e.g. "string" for text/plain, "Foo" for an `application/json` $ref).
+    pub body_type: String,
+    /// Whether the body is required.
+    pub required: bool,
+}
+
+/// Endpoint information extracted from OpenAPI spec.
+#[derive(Debug, Clone)]
+pub struct Endpoint {
+    /// HTTP method (GET, POST, etc.)
+    pub method: String,
+    /// Path template (e.g., "/blocks/{hash}")
+    pub path: String,
+    /// Operation ID (e.g., "getBlockByHash")
+    pub operation_id: Option<String>,
+    /// Short summary
+    pub summary: Option<String>,
+    /// Detailed description
+    pub description: Option<String>,
+    /// Path parameters
+    pub path_params: Vec<Parameter>,
+    /// Query parameters
+    pub query_params: Vec<Parameter>,
+    /// Request body, if any (POST/PUT/PATCH).
+    pub request_body: Option<RequestBody>,
+    /// Body kind for the 200 response.
+    pub response_kind: ResponseKind,
+    /// Whether this endpoint is deprecated
+    pub deprecated: bool,
+    /// Whether this endpoint supports CSV format (text/csv content type)
+    pub supports_csv: bool,
+}
+
+impl Endpoint {
+    /// Returns true if this endpoint should be included in client generation.
+    /// Non-deprecated GET and POST endpoints are included.
+    pub fn should_generate(&self) -> bool {
+        !self.deprecated && (self.method == "GET" || self.method == "POST")
+    }
+
+    /// Returns true if this endpoint returns JSON.
+    pub fn returns_json(&self) -> bool {
+        matches!(self.response_kind, ResponseKind::Json(_))
+    }
+
+    /// Returns true if this endpoint returns binary data (application/octet-stream).
+    pub fn returns_binary(&self) -> bool {
+        matches!(self.response_kind, ResponseKind::Binary)
+    }
+
+    /// Returns true if this endpoint returns plain text (typed or opaque).
+    pub fn returns_text(&self) -> bool {
+        matches!(self.response_kind, ResponseKind::Text(_))
+    }
+
+    /// Schema name attached to the response, if any.
+    pub fn schema_name(&self) -> Option<&str> {
+        self.response_kind.schema_name()
+    }
+
+    /// Returns the operation ID or generates one from the path.
+    /// The returned string uses the raw case from the spec (typically camelCase).
+    pub fn operation_name(&self) -> String {
+        if let Some(op_id) = &self.operation_id {
+            return op_id.clone();
+        }
+        let mut parts: Vec<String> = Vec::new();
+        let mut prev_segment = "";
+
+        for segment in self.path.split('/').filter(|s| !s.is_empty()) {
+            if segment == "api" {
+                continue;
+            }
+            if let Some(param) = segment.strip_prefix('{').and_then(|s| s.strip_suffix('}')) {
+                let prev_normalized = prev_segment.replace('-', "_");
+                if !prev_normalized.ends_with(param) {
+                    parts.push(format!("by_{}", param));
+                }
+            } else {
+                let normalized = segment.replace('-', "_");
+                parts.push(normalized);
+                prev_segment = segment;
+            }
+        }
+        format!("get_{}", parts.join("_"))
+    }
+}

crates/brk_bindgen/src/openapi/mod.rs

@@ -0,0 +1,372 @@
+mod endpoint;
+mod parameter;
+mod response_kind;
+mod text_schema;
+
+pub use endpoint::{Endpoint, RequestBody};
+pub use parameter::Parameter;
+pub use response_kind::ResponseKind;
+pub use text_schema::TextSchema;
+
+use std::{collections::BTreeMap, io};
+
+use crate::ref_to_type_name;
+use oas3::Spec;
+use oas3::spec::{
+    ObjectOrReference, ObjectSchema, Operation, ParameterIn, PathItem, Schema, SchemaType,
+    SchemaTypeSet,
+};
+use serde_json::Value;
+
+/// Type schema extracted from OpenAPI components
+pub type TypeSchemas = BTreeMap<String, Value>;
+
+/// Parse OpenAPI spec from JSON string
+///
+/// Pre-processes the JSON to handle oas3 limitations:
+/// - Removes unsupported siblings from `$ref` objects (oas3 only supports `summary` and `description`)
+pub fn parse_openapi_json(json: &str) -> io::Result<Spec> {
+    let mut value: Value =
+        serde_json::from_str(json).map_err(|e| io::Error::new(io::ErrorKind::InvalidData, e))?;
+
+    // Clean up for oas3 compatibility
+    clean_for_oas3(&mut value);
+
+    let cleaned_json =
+        serde_json::to_string(&value).map_err(|e| io::Error::new(io::ErrorKind::InvalidData, e))?;
+
+    oas3::from_json(&cleaned_json).map_err(|e| io::Error::new(io::ErrorKind::InvalidData, e))
+}
+
+/// Extract type schemas from OpenAPI JSON
+pub fn extract_schemas(json: &str) -> TypeSchemas {
+    let Ok(value) = serde_json::from_str::<Value>(json) else {
+        return BTreeMap::new();
+    };
+
+    value
+        .get("components")
+        .and_then(|c| c.get("schemas"))
+        .and_then(|s| s.as_object())
+        .map(|schemas| {
+            schemas
+                .iter()
+                .map(|(name, schema)| (name.clone(), schema.clone()))
+                .collect()
+        })
+        .unwrap_or_default()
+}
+
+/// Clean up OpenAPI spec for oas3 compatibility.
+/// - Removes unsupported siblings from $ref objects (oas3 only supports summary and description)
+/// - Converts boolean schemas to object schemas (oas3 doesn't handle `"schema": true`)
+fn clean_for_oas3(value: &mut Value) {
+    match value {
+        Value::Object(map) => {
+            // Handle $ref with unsupported siblings
+            if map.contains_key("$ref") {
+                map.retain(|k, _| k == "$ref" || k == "summary" || k == "description");
+            } else {
+                // Convert boolean schemas to empty object schemas
+                if let Some(schema) = map.get_mut("schema")
+                    && schema.is_boolean()
+                {
+                    *schema = Value::Object(serde_json::Map::new());
+                }
+                for v in map.values_mut() {
+                    clean_for_oas3(v);
+                }
+            }
+        }
+        Value::Array(arr) => {
+            for v in arr {
+                clean_for_oas3(v);
+            }
+        }
+        _ => {}
+    }
+}
+
+/// Extract all endpoints from OpenAPI spec
+pub fn extract_endpoints(spec: &Spec) -> Vec<Endpoint> {
+    let mut endpoints = Vec::new();
+
+    let Some(paths) = &spec.paths else {
+        return endpoints;
+    };
+
+    for (path, path_item) in paths {
+        for (method, operation) in get_operations(path_item) {
+            if let Some(endpoint) = extract_endpoint(path, method, operation, spec) {
+                endpoints.push(endpoint);
+            }
+        }
+    }
+
+    endpoints
+}
+
+fn get_operations(path_item: &PathItem) -> Vec<(&'static str, &Operation)> {
+    [
+        ("GET", &path_item.get),
+        ("POST", &path_item.post),
+        ("PUT", &path_item.put),
+        ("DELETE", &path_item.delete),
+        ("PATCH", &path_item.patch),
+    ]
+    .into_iter()
+    .filter_map(|(method, op)| op.as_ref().map(|o| (method, o)))
+    .collect()
+}
+
+fn extract_endpoint(
+    path: &str,
+    method: &str,
+    operation: &Operation,
+    spec: &Spec,
+) -> Option<Endpoint> {
+    let path_params = extract_path_parameters(path, operation);
+    let query_params = extract_parameters(operation, ParameterIn::Query);
+
+    let response_kind = extract_response_kind(operation, spec);
+    let request_body = extract_request_body(operation);
+    let supports_csv = check_csv_support(operation);
+
+    Some(Endpoint {
+        method: method.to_string(),
+        path: path.to_string(),
+        operation_id: operation.operation_id.clone(),
+        summary: operation.summary.clone(),
+        description: operation.description.clone(),
+        path_params,
+        query_params,
+        request_body,
+        response_kind,
+        deprecated: operation.deprecated.unwrap_or(false),
+        supports_csv,
+    })
+}
+
+/// Extract the request body shape, if any.
+/// Prefers `text/plain` (string) over `application/json` (typed).
+fn extract_request_body(operation: &Operation) -> Option<RequestBody> {
+    let req = operation.request_body.as_ref()?;
+    let req = match req {
+        ObjectOrReference::Object(rb) => rb,
+        ObjectOrReference::Ref { .. } => return None,
+    };
+
+    let body_type = if req.content.contains_key("text/plain; charset=utf-8")
+        || req.content.contains_key("text/plain")
+    {
+        "string".to_string()
+    } else if let Some(content) = req.content.get("application/json") {
+        schema_name_from_content(content).unwrap_or_else(|| "Object".to_string())
+    } else {
+        "string".to_string()
+    };
+
+    Some(RequestBody {
+        body_type,
+        required: req.required.unwrap_or(false),
+    })
+}
+
+/// Check if the endpoint supports CSV format (has text/csv in 200 response content types).
+fn check_csv_support(operation: &Operation) -> bool {
+    let Some(responses) = operation.responses.as_ref() else {
+        return false;
+    };
+    let Some(response) = responses.get("200") else {
+        return false;
+    };
+    match response {
+        ObjectOrReference::Object(response) => response.content.contains_key("text/csv"),
+        ObjectOrReference::Ref { .. } => false,
+    }
+}
+
+/// Extract path parameters in the order they appear in the path URL.
+fn extract_path_parameters(path: &str, operation: &Operation) -> Vec<Parameter> {
+    // Extract parameter names from the path in order (e.g., "/api/series/{series}/{index}" -> ["series", "index"])
+    let path_order: Vec<&str> = path
+        .split('/')
+        .filter_map(|segment| segment.strip_prefix('{').and_then(|s| s.strip_suffix('}')))
+        .collect();
+
+    // Get all path parameters from the operation
+    let params = extract_parameters(operation, ParameterIn::Path);
+
+    // Sort by position in the path
+    let mut sorted_params: Vec<Parameter> = params;
+    sorted_params.sort_by_key(|p| {
+        path_order
+            .iter()
+            .position(|&name| name == p.name)
+            .unwrap_or(usize::MAX)
+    });
+
+    sorted_params
+}
+
+fn extract_parameters(operation: &Operation, location: ParameterIn) -> Vec<Parameter> {
+    operation
+        .parameters
+        .iter()
+        .filter_map(|p| match p {
+            ObjectOrReference::Object(param) if param.location == location => {
+                let param_type = param
+                    .schema
+                    .as_ref()
+                    .and_then(schema_type_from_schema)
+                    .unwrap_or_else(|| "string".to_string());
+                Some(Parameter {
+                    name: param.name.clone(),
+                    required: param.required.unwrap_or(false),
+                    param_type,
+                    description: param.description.clone(),
+                })
+            }
+            _ => None,
+        })
+        .collect()
+}
+
+fn extract_response_kind(operation: &Operation, spec: &Spec) -> ResponseKind {
+    let response = operation
+        .responses
+        .as_ref()
+        .and_then(|r| r.get("200"))
+        .and_then(|r| match r {
+            ObjectOrReference::Object(o) => Some(o),
+            ObjectOrReference::Ref { .. } => None,
+        });
+    let Some(response) = response else {
+        return ResponseKind::Text(None);
+    };
+
+    if response.content.contains_key("application/octet-stream") {
+        return ResponseKind::Binary;
+    }
+    if let Some(content) = response.content.get("application/json") {
+        return ResponseKind::Json(
+            schema_name_from_content(content).unwrap_or_else(|| "*".to_string()),
+        );
+    }
+    if let Some(content) = response.content.get("text/plain; charset=utf-8") {
+        let schema = schema_name_from_content(content).map(|name| {
+            let is_numeric = is_numeric_schema(spec, &name);
+            TextSchema { name, is_numeric }
+        });
+        return ResponseKind::Text(schema);
+    }
+    ResponseKind::Text(None)
+}
+
+fn schema_name_from_content(content: &oas3::spec::MediaType) -> Option<String> {
+    schema_type_from_schema(content.schema.as_ref()?)
+}
+
+/// Resolves `name` against `components.schemas` and reports whether the
+/// underlying primitive is `integer` or `number`.
+fn is_numeric_schema(spec: &Spec, name: &str) -> bool {
+    let Some(components) = spec.components.as_ref() else {
+        return false;
+    };
+    let Some(Schema::Object(obj_or_ref)) = components.schemas.get(name) else {
+        return false;
+    };
+    let ObjectOrReference::Object(schema) = obj_or_ref.as_ref() else {
+        return false;
+    };
+    matches!(
+        schema.schema_type.as_ref(),
+        Some(SchemaTypeSet::Single(
+            SchemaType::Integer | SchemaType::Number
+        ))
+    )
+}
+
+fn schema_type_from_schema(schema: &Schema) -> Option<String> {
+    match schema {
+        Schema::Boolean(_) => Some("boolean".to_string()),
+        Schema::Object(obj_or_ref) => match obj_or_ref.as_ref() {
+            ObjectOrReference::Object(obj_schema) => schema_to_type_name(obj_schema),
+            ObjectOrReference::Ref { ref_path, .. } => {
+                // Return the type name as-is (e.g., "Height", "Address")
+                // These should have definitions generated from schemas
+                ref_to_type_name(ref_path).map(|s| s.to_string())
+            }
+        },
+    }
+}
+
+fn schema_to_type_name(schema: &ObjectSchema) -> Option<String> {
+    if let Some(schema_type) = schema.schema_type.as_ref() {
+        return match schema_type {
+            SchemaTypeSet::Single(t) => single_type_to_name(t, schema),
+            SchemaTypeSet::Multiple(types) => {
+                // For nullable types like ["integer", "null"], return the non-null type
+                types
+                    .iter()
+                    .find(|t| !matches!(t, SchemaType::Null))
+                    .and_then(|t| single_type_to_name(t, schema))
+                    .or(Some("*".to_string()))
+            }
+        };
+    }
+
+    // Handle anyOf/oneOf unions (e.g., Option<RangeIndex> → anyOf: [$ref, null])
+    let variants = if !schema.any_of.is_empty() {
+        &schema.any_of
+    } else if !schema.one_of.is_empty() {
+        &schema.one_of
+    } else {
+        return None;
+    };
+
+    let types: Vec<String> = variants
+        .iter()
+        .filter_map(|v| match v {
+            Schema::Boolean(_) => None,
+            Schema::Object(obj_or_ref) => match obj_or_ref.as_ref() {
+                ObjectOrReference::Ref { ref_path, .. } => {
+                    ref_to_type_name(ref_path).map(|s| s.to_string())
+                }
+                ObjectOrReference::Object(obj) => {
+                    if matches!(
+                        obj.schema_type.as_ref(),
+                        Some(SchemaTypeSet::Single(SchemaType::Null))
+                    ) {
+                        return None;
+                    }
+                    schema_to_type_name(obj)
+                }
+            },
+        })
+        .collect();
+
+    match types.len() {
+        0 => None,
+        1 => Some(types.into_iter().next().unwrap()),
+        _ => Some(types.join(" | ")),
+    }
+}
+
+fn single_type_to_name(t: &SchemaType, schema: &ObjectSchema) -> Option<String> {
+    match t {
+        SchemaType::String => Some("string".to_string()),
+        SchemaType::Number => Some("number".to_string()),
+        SchemaType::Integer => Some("integer".to_string()),
+        SchemaType::Boolean => Some("boolean".to_string()),
+        SchemaType::Array => {
+            let inner = match &schema.items {
+                Some(boxed_schema) => schema_type_from_schema(boxed_schema),
+                None => Some("*".to_string()),
+            };
+            inner.map(|t| format!("{}[]", t))
+        }
+        SchemaType::Object => Some("Object".to_string()),
+        SchemaType::Null => Some("null".to_string()),
+    }
+}

crates/brk_bindgen/src/openapi/parameter.rs

@@ -0,0 +1,8 @@
+/// Parameter information.
+#[derive(Debug, Clone)]
+pub struct Parameter {
+    pub name: String,
+    pub required: bool,
+    pub param_type: String,
+    pub description: Option<String>,
+}

crates/brk_bindgen/src/openapi/response_kind.rs

@@ -0,0 +1,29 @@
+use crate::openapi::TextSchema;
+
+/// 200-response body shape.
+#[derive(Debug, Clone)]
+pub enum ResponseKind {
+    /// JSON body, schema named (e.g. "Block").
+    Json(String),
+    /// `text/plain` body. `Some(schema)` carries a typed shape (e.g. "Height", "Hex");
+    /// `None` is the escape hatch for opaque text.
+    Text(Option<TextSchema>),
+    /// `application/octet-stream`.
+    Binary,
+}
+
+impl ResponseKind {
+    /// Schema name, if the body is named (Json or typed Text).
+    pub fn schema_name(&self) -> Option<&str> {
+        match self {
+            Self::Json(s) => Some(s.as_str()),
+            Self::Text(Some(t)) => Some(t.name.as_str()),
+            _ => None,
+        }
+    }
+
+    /// True when a typed text body needs numeric parsing (`int(...)` etc.).
+    pub fn text_is_numeric(&self) -> bool {
+        matches!(self, Self::Text(Some(t)) if t.is_numeric)
+    }
+}

crates/brk_bindgen/src/openapi/text_schema.rs

@@ -0,0 +1,8 @@
+/// Schema metadata for a typed `text/plain` response.
+#[derive(Debug, Clone)]
+pub struct TextSchema {
+    /// Schema name, e.g. "Height", "Hex".
+    pub name: String,
+    /// True when the underlying primitive is `integer`/`number` (body needs numeric parsing).
+    pub is_numeric: bool,
+}

crates/brk_bindgen/src/syntax.rs

@@ -0,0 +1,122 @@
+//! Language-specific syntax traits for code generation.
+//!
+//! This module defines the `LanguageSyntax` trait that abstracts over
+//! language-specific code generation patterns, allowing shared generation
+//! logic to work across Python, JavaScript, and Rust backends.
+
+use crate::GenericSyntax;
+
+/// Language-specific syntax for code generation.
+///
+/// Implementations of this trait provide the language-specific formatting
+/// for generated client code. This allows the core generation logic to be
+/// written once and reused across all supported languages.
+pub trait LanguageSyntax {
+    /// Convert a field name to the language's naming convention.
+    ///
+    /// - Python/Rust: `snake_case`
+    /// - JavaScript: `camelCase`
+    fn field_name(&self, name: &str) -> String;
+
+    /// Format an interpolated path expression.
+    ///
+    /// # Arguments
+    /// * `base_var` - The variable name to interpolate (e.g., "acc", "base_path")
+    /// * `suffix` - The suffix to append (e.g., "_field_name")
+    ///
+    /// # Returns
+    /// - Python: `f'{acc}_suffix'`
+    /// - JavaScript: `` `${acc}_suffix` ``
+    /// - Rust: `format!("{acc}_suffix")`
+    fn path_expr(&self, base_var: &str, suffix: &str) -> String;
+
+    /// Format a suffix mode expression: `_m(acc, relative)`.
+    ///
+    /// Suffix mode appends the relative name to the accumulator.
+    /// - If relative is empty, returns just acc (identity)
+    /// - Otherwise: `{acc}_{relative}` or `{relative}` if acc is empty
+    ///
+    /// # Arguments
+    /// * `acc_var` - The accumulator variable name (e.g., "acc")
+    /// * `relative` - The relative name to append (e.g., "max_cost_basis")
+    fn suffix_expr(&self, acc_var: &str, relative: &str) -> String;
+
+    /// Format a prefix mode expression: `_p(prefix, acc)`.
+    ///
+    /// Prefix mode prepends the prefix to the accumulator.
+    /// - If prefix is empty, returns just acc (identity)
+    /// - Otherwise: `{prefix}{acc}` (prefix includes trailing underscore)
+    ///
+    /// # Arguments
+    /// * `prefix` - The prefix to prepend (e.g., "cumulative_")
+    /// * `acc_var` - The accumulator variable name (e.g., "acc")
+    fn prefix_expr(&self, prefix: &str, acc_var: &str) -> String;
+
+    /// Generate a constructor call for patterns and accessors.
+    ///
+    /// - Python: `TypeName(client, path)`
+    /// - JavaScript: `createTypeName(client, path)`
+    /// - Rust: `TypeName::new(client.clone(), path)`
+    fn constructor(&self, type_name: &str, path_expr: &str) -> String;
+
+    /// Generate a field initialization line.
+    ///
+    /// # Arguments
+    /// * `indent` - The indentation string
+    /// * `name` - The field name (already converted to language convention)
+    /// * `type_ann` - The type annotation (may be ignored by some languages)
+    /// * `value` - The initialization value/expression
+    ///
+    /// # Returns
+    /// - Python: `{indent}self.{name}: {type_ann} = {value}`
+    /// - JavaScript: `{indent}{name}: {value},`
+    /// - Rust: `{indent}{name}: {value},`
+    fn field_init(&self, indent: &str, name: &str, type_ann: &str, value: &str) -> String;
+
+    /// Get the generic type syntax for this language.
+    ///
+    /// - Python: `[T]` with default `Any`
+    /// - JavaScript: `<T>` with default `unknown`
+    /// - Rust: `<T>` with default `_`
+    fn generic_syntax(&self) -> GenericSyntax;
+
+    /// Format a string literal.
+    ///
+    /// - Python/JavaScript: `'value'` (single quotes)
+    /// - Rust: `"value"` (double quotes)
+    fn string_literal(&self, value: &str) -> String;
+
+    /// Get the constructor name/prefix for a type.
+    ///
+    /// - Python: `TypeName`
+    /// - JavaScript: `createTypeName`
+    /// - Rust: `TypeName::new`
+    fn constructor_name(&self, type_name: &str) -> String;
+
+    /// Return a variable as an owned value expression.
+    ///
+    /// - Rust: `var.clone()` (String needs explicit cloning)
+    /// - JavaScript/Python: `var` (no ownership)
+    fn owned_expr(&self, var: &str) -> String {
+        var.to_string()
+    }
+
+    /// Format a discriminator argument for passing to a templated child.
+    ///
+    /// Returns an expression computing the disc value from a template.
+    /// - `"pct99"` (static) → `'pct99'` (JS) / `"pct99".to_string()` (Rust)
+    /// - `""` (empty) → `disc` (pass parent's disc through)
+    /// - `"p1sd{disc}"` (suffix) → `_m('p1sd', disc)` (composed)
+    /// - `"ratio_{disc}_bps"` (embedded) → `` `ratio_${disc}_bps` `` (template literal)
+    fn disc_arg_expr(&self, template: &str) -> String;
+
+    /// Format a templated mode expression: substitute `{disc}` at runtime.
+    ///
+    /// The template contains `{disc}` placeholder. The generated code should
+    /// construct `_m(acc, template_with_disc_substituted)` at runtime.
+    ///
+    /// # Arguments
+    /// * `acc_var` - The accumulator variable (e.g., "acc")
+    /// * `template` - Template like `"ratio_{disc}_bps"` or `"{disc}"`
+    fn template_expr(&self, acc_var: &str, template: &str) -> String;
+}

crates/brk_bindgen/src/types/case.rs

@@ -0,0 +1,93 @@
+use brk_types::Index;
+
+/// Convert a string to PascalCase (e.g., "fee_rate" -> "FeeRate").
+pub fn to_pascal_case(s: &str) -> String {
+    s.replace('-', "_")
+        .split('_')
+        .map(|word| {
+            let mut chars = word.chars();
+            match chars.next() {
+                None => String::new(),
+                Some(first) => first.to_uppercase().collect::<String>() + chars.as_str(),
+            }
+        })
+        .collect()
+}
+
+/// Convert a string to snake_case (no keyword escaping — backends handle that).
+pub fn to_snake_case(s: &str) -> String {
+    let sanitized = s.to_lowercase().replace('-', "_");
+
+    if sanitized.chars().next().is_some_and(|c| c.is_ascii_digit()) {
+        format!("_{}", sanitized)
+    } else {
+        sanitized
+    }
+}
+
+/// Escape Rust reserved keywords with `_` suffix (consistent with Python).
+pub fn escape_rust_keyword(name: &str) -> String {
+    match name {
+        "type" | "const" | "static" | "match" | "if" | "else" | "loop" | "while" | "for"
+        | "break" | "continue" | "return" | "fn" | "let" | "mut" | "ref" | "self" | "super"
+        | "mod" | "use" | "pub" | "crate" | "extern" | "impl" | "trait" | "struct" | "enum"
+        | "where" | "async" | "await" | "dyn" | "move" => format!("{}_", name),
+        _ => name.to_string(),
+    }
+}
+
+/// Convert a string to camelCase (e.g., "fee_rate" -> "feeRate").
+pub fn to_camel_case(s: &str) -> String {
+    let pascal = to_pascal_case(s);
+    let mut chars = pascal.chars();
+
+    let result = match chars.next() {
+        None => String::new(),
+        Some(first) => first.to_lowercase().collect::<String>() + chars.as_str(),
+    };
+
+    // Prefix with _ if starts with digit
+    if result.chars().next().is_some_and(|c| c.is_ascii_digit()) {
+        format!("_{}", result)
+    } else {
+        result
+    }
+}
+
+/// Convert an Index to a snake_case field name (e.g., Day1 -> day1).
+pub fn index_to_field_name(index: &Index) -> String {
+    to_snake_case(index.name())
+}
+
+/// Generate a child type/struct/class name (e.g., ParentName + child_name -> ParentName_ChildName).
+pub fn child_type_name(parent: &str, child: &str) -> String {
+    format!("{}_{}", parent, to_pascal_case(child))
+}
+
+/// Escape Python reserved keywords by appending an underscore.
+/// Also prefixes names starting with digits with an underscore.
+pub fn escape_python_keyword(name: &str) -> String {
+    const PYTHON_KEYWORDS: &[&str] = &[
+        "False", "None", "True", "and", "as", "assert", "async", "await", "break", "class",
+        "continue", "def", "del", "elif", "else", "except", "finally", "for", "from", "global",
+        "if", "import", "in", "is", "lambda", "nonlocal", "not", "or", "pass", "raise", "return",
+        "try", "while", "with", "yield",
+    ];
+
+    // Strip characters invalid in identifiers (e.g. `[]` from `txId[]`)
+    let name = name.replace(['[', ']'], "");
+
+    // Prefix with underscore if starts with digit
+    let name = if name.starts_with(|c: char| c.is_ascii_digit()) {
+        format!("_{}", name)
+    } else {
+        name.to_string()
+    };
+
+    // Append underscore if it's a keyword
+    if PYTHON_KEYWORDS.contains(&name.as_str()) {
+        format!("{}_", name)
+    } else {
+        name
+    }
+}

crates/brk_bindgen/src/types/metadata.rs

@@ -0,0 +1,158 @@
+//! Client metadata extracted from brk_query.
+
+use std::collections::{BTreeMap, BTreeSet};
+
+use brk_query::Vecs;
+use brk_types::{Index, SeriesLeafWithSchema};
+
+use super::{GenericSyntax, IndexSetPattern, PatternField, StructuralPattern, extract_inner_type};
+use crate::{PatternBaseResult, analysis};
+
+/// Metadata extracted from brk_query for client generation.
+#[derive(Debug)]
+pub struct ClientMetadata {
+    /// The catalog tree structure (with schemas in leaves)
+    pub catalog: brk_types::TreeNode,
+    /// Structural patterns - tree node shapes that repeat
+    pub structural_patterns: Vec<StructuralPattern>,
+    /// Index set patterns - sets of indexes that appear together on series
+    pub index_set_patterns: Vec<IndexSetPattern>,
+    /// Maps field signatures to pattern names (merged from concrete instances + pattern definitions)
+    pattern_lookup: BTreeMap<Vec<PatternField>, String>,
+    /// Maps concrete field signatures to their type parameter (for generic patterns)
+    concrete_to_type_param: BTreeMap<Vec<PatternField>, String>,
+    /// Maps tree paths to their computed PatternBaseResult
+    node_bases: BTreeMap<String, PatternBaseResult>,
+}
+
+impl ClientMetadata {
+    /// Extract metadata from brk_query::Vecs.
+    pub fn from_vecs(vecs: &Vecs) -> Self {
+        Self::from_catalog(vecs.catalog().clone())
+    }
+
+    /// Extract metadata from a catalog TreeNode directly.
+    pub fn from_catalog(catalog: brk_types::TreeNode) -> Self {
+        let (structural_patterns, concrete_to_pattern, concrete_to_type_param, node_bases) =
+            analysis::detect_structural_patterns(&catalog);
+        let index_set_patterns = analysis::detect_index_patterns(&catalog);
+
+        // Build merged pattern lookup: concrete instances + pattern definitions
+        let mut pattern_lookup = concrete_to_pattern;
+        for p in &structural_patterns {
+            pattern_lookup.insert(p.fields.clone(), p.name.clone());
+        }
+
+        ClientMetadata {
+            catalog,
+            structural_patterns,
+            index_set_patterns,
+            pattern_lookup,
+            concrete_to_type_param,
+            node_bases,
+        }
+    }
+
+    /// Find an index set pattern that matches the given indexes.
+    pub fn find_index_set_pattern(&self, indexes: &BTreeSet<Index>) -> Option<&IndexSetPattern> {
+        self.index_set_patterns
+            .iter()
+            .find(|p| &p.indexes == indexes)
+    }
+
+    /// Find a pattern by name.
+    pub fn find_pattern(&self, name: &str) -> Option<&StructuralPattern> {
+        self.structural_patterns.iter().find(|p| p.name == name)
+    }
+
+    /// Check if a pattern is fully parameterizable (recursively).
+    /// Returns false if the pattern or any nested branch pattern has no mode.
+    pub fn is_parameterizable(&self, name: &str) -> bool {
+        self.find_pattern(name).is_some_and(|p| {
+            p.is_parameterizable()
+                && p.fields.iter().all(|f| {
+                    !f.is_branch()
+                        || self.find_pattern(&f.rust_type).is_none()
+                        || self.is_parameterizable(&f.rust_type)
+                })
+        })
+    }
+
+    /// Find a pattern by its concrete fields.
+    pub fn find_pattern_by_fields(&self, fields: &[PatternField]) -> Option<&StructuralPattern> {
+        self.pattern_lookup
+            .get(fields)
+            .and_then(|name| self.find_pattern(name))
+    }
+
+    /// Get the type parameter for a generic pattern given its concrete fields.
+    pub fn get_type_param(&self, fields: &[PatternField]) -> Option<&String> {
+        self.concrete_to_type_param.get(fields)
+    }
+
+    /// Get the pre-computed pattern lookup map.
+    pub fn pattern_lookup(&self) -> &BTreeMap<Vec<PatternField>, String> {
+        &self.pattern_lookup
+    }
+
+    /// Get the pre-computed PatternBaseResult for a tree path.
+    pub fn get_node_base(&self, path: &str) -> Option<&PatternBaseResult> {
+        self.node_bases.get(path)
+    }
+
+    /// Generate type annotation for a field with language-specific syntax.
+    pub fn field_type_annotation(
+        &self,
+        field: &PatternField,
+        is_generic: bool,
+        generic_value_type: Option<&str>,
+        syntax: GenericSyntax,
+    ) -> String {
+        // Pattern type — single lookup instead of is_pattern_type + is_pattern_generic
+        if let Some(pattern) = self.find_pattern(&field.rust_type) {
+            if pattern.is_generic {
+                let type_param = field
+                    .type_param
+                    .as_deref()
+                    .or(generic_value_type)
+                    .unwrap_or(if is_generic { "T" } else { syntax.default_type });
+                return syntax.wrap(&field.rust_type, type_param);
+            }
+            return field.rust_type.clone();
+        }
+
+        // Branch type (non-pattern)
+        if field.is_branch() {
+            return field.rust_type.clone();
+        }
+
+        // Leaf type
+        let value_type = if is_generic && field.rust_type == "T" {
+            "T".to_string()
+        } else {
+            extract_inner_type(&field.rust_type)
+        };
+        if let Some(accessor) = self.find_index_set_pattern(&field.indexes) {
+            syntax.wrap(&accessor.name, &value_type)
+        } else {
+            syntax.wrap("SeriesNode", &value_type)
+        }
+    }
+
+    /// Generate type annotation for a leaf node with language-specific syntax.
+    ///
+    /// This is a simpler version of `field_type_annotation` that works directly
+    /// with a `SeriesLeafWithSchema` node instead of a `PatternField`.
+    pub fn field_type_annotation_from_leaf(
+        &self,
+        leaf: &SeriesLeafWithSchema,
+        syntax: GenericSyntax,
+    ) -> String {
+        let value_type = leaf.kind().to_string();
+        if let Some(accessor) = self.find_index_set_pattern(leaf.indexes()) {
+            syntax.wrap(&accessor.name, &value_type)
+        } else {
+            syntax.wrap("SeriesNode", &value_type)
+        }
+    }
+}

crates/brk_bindgen/src/types/mod.rs

@@ -0,0 +1,68 @@
+//! Core types for client generation.
+
+mod case;
+mod metadata;
+mod positions;
+mod schema;
+mod structs;
+
+pub use case::*;
+pub use metadata::*;
+pub use positions::*;
+pub use schema::*;
+pub use structs::*;
+
+/// Language-specific syntax for generic type annotations.
+#[derive(Clone, Copy)]
+pub struct GenericSyntax {
+    pub open: char,
+    pub close: char,
+    pub default_type: &'static str,
+}
+
+impl GenericSyntax {
+    pub const PYTHON: Self = Self {
+        open: '[',
+        close: ']',
+        default_type: "Any",
+    };
+    pub const JAVASCRIPT: Self = Self {
+        open: '<',
+        close: '>',
+        default_type: "unknown",
+    };
+    pub const RUST: Self = Self {
+        open: '<',
+        close: '>',
+        default_type: "_",
+    };
+
+    pub fn wrap(&self, name: &str, type_param: &str) -> String {
+        // Convert the type_param from Rust syntax to target syntax
+        let converted = self.convert(type_param);
+        format!("{}{}{}{}", name, self.open, converted, self.close)
+    }
+
+    /// Convert a type string from Rust generic syntax to target language syntax.
+    ///
+    /// For Python, wrapper newtypes like `Close<Cents>` are flattened to just `Cents`
+    /// because Python type aliases can't be parameterized. This matches JS behavior.
+    pub fn convert(&self, type_str: &str) -> String {
+        // Flatten nested generics to innermost type (e.g., Close<Cents> -> Cents)
+        // This is needed because wrapper types like Close, Open, High, Low are
+        // just type aliases in generated code, not actual generic classes.
+        extract_inner_type_recursive(type_str)
+    }
+}
+
+/// Extract the innermost type from nested generics.
+/// E.g., `Close<Cents>` -> `Cents`, `Foo<Bar<Baz>>` -> `Baz`
+fn extract_inner_type_recursive(type_str: &str) -> String {
+    if let Some(start) = type_str.find('<')
+        && let Some(end) = type_str.rfind('>')
+    {
+        let inner = &type_str[start + 1..end];
+        return extract_inner_type_recursive(inner);
+    }
+    type_str.to_string()
+}

crates/brk_bindgen/src/types/positions.rs

@@ -0,0 +1,34 @@
+//! Pattern mode and field parts for series name reconstruction.
+//!
+//! Patterns are either suffix mode or prefix mode:
+//! - Suffix mode: `_m(acc, relative)` → `acc_relative` or just `relative` if acc empty
+//! - Prefix mode: `_p(prefix, acc)` → `prefix_acc` or just `acc` if prefix empty
+
+use std::collections::BTreeMap;
+
+/// How a pattern constructs series names from the accumulator.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum PatternMode {
+    /// Fields append their relative name to acc.
+    /// Formula: `_m(acc, relative)` → `{acc}_{relative}` or `{relative}` if acc empty
+    /// Example: `_m("lth", "max_cost_basis")` → `"lth_max_cost_basis"`
+    Suffix {
+        /// Maps field name to its relative name (full series name when acc = "")
+        relatives: BTreeMap<String, String>,
+    },
+    /// Fields prepend their prefix to acc.
+    /// Formula: `_p(prefix, acc)` → `{prefix}_{acc}` or `{acc}` if prefix empty
+    /// Example: `_p("cumulative", "lth_realized_loss")` → `"cumulative_lth_realized_loss"`
+    Prefix {
+        /// Maps field name to its prefix (empty string for identity)
+        prefixes: BTreeMap<String, String>,
+    },
+    /// Fields construct series names using a template with a discriminator placeholder.
+    /// Factory takes two params: `acc` (base) and `disc` (discriminator).
+    /// Formula: `_m(acc, template.replace("{disc}", disc))`
+    /// Example: template `"ratio_{disc}_bps"` with disc `"pct99"` → `_m(acc, "ratio_pct99_bps")`
+    Templated {
+        /// Maps field name to its template string containing `{disc}` placeholder
+        templates: BTreeMap<String, String>,
+    },
+}

crates/brk_bindgen/src/types/schema.rs

@@ -0,0 +1,43 @@
+use serde_json::Value;
+
+/// Unwrap allOf with a single element, returning the inner schema.
+/// Schemars uses allOf for composition, but often with just one $ref.
+pub fn unwrap_allof(schema: &Value) -> &Value {
+    if let Some(all_of) = schema.get("allOf").and_then(|v| v.as_array())
+        && all_of.len() == 1
+    {
+        return &all_of[0];
+    }
+    schema
+}
+
+/// Extract inner type from a wrapper generic like `Close<Dollars>` -> `Dollars`.
+/// Also handles malformed types like `Dollars>` (from vecdb's short_type_name).
+pub fn extract_inner_type(type_str: &str) -> String {
+    // Handle proper generic wrappers like `Close<Dollars>` -> `Dollars`
+    if let Some(start) = type_str.find('<')
+        && let Some(end) = type_str.rfind('>')
+        && start < end
+    {
+        return type_str[start + 1..end].to_string();
+    }
+    // Handle malformed types like `Dollars>` (trailing > without <)
+    if type_str.ends_with('>') && !type_str.contains('<') {
+        return type_str.trim_end_matches('>').to_string();
+    }
+    type_str.to_string()
+}
+
+/// Extract type name from a JSON Schema $ref path.
+/// E.g., "#/definitions/MyType" -> "MyType", "#/$defs/Foo" -> "Foo"
+pub fn ref_to_type_name(ref_path: &str) -> Option<&str> {
+    ref_path.rsplit('/').next()
+}
+
+/// Get union variants from anyOf or oneOf schema.
+pub fn get_union_variants(schema: &Value) -> Option<&Vec<Value>> {
+    schema
+        .get("anyOf")
+        .or_else(|| schema.get("oneOf"))
+        .and_then(|v| v.as_array())
+}

crates/brk_bindgen/src/types/structs.rs

@@ -0,0 +1,188 @@
+//! Structural pattern and field types.
+
+use std::collections::{BTreeMap, BTreeSet};
+
+use brk_types::Index;
+
+use super::PatternMode;
+
+/// A pattern of indexes that appear together on multiple series.
+#[derive(Debug, Clone)]
+pub struct IndexSetPattern {
+    /// Pattern name (e.g., "DateHeightIndexes")
+    pub name: String,
+    /// The set of indexes
+    pub indexes: BTreeSet<Index>,
+}
+
+/// A structural pattern - a branch structure that appears multiple times.
+#[derive(Debug, Clone)]
+pub struct StructuralPattern {
+    /// Pattern name
+    pub name: String,
+    /// Ordered list of child fields
+    pub fields: Vec<PatternField>,
+    /// How fields construct series names from acc (None = not parameterizable)
+    pub mode: Option<PatternMode>,
+    /// If true, all leaf fields use a type parameter T
+    pub is_generic: bool,
+}
+
+impl StructuralPattern {
+    /// Returns true if this pattern can be parameterized with an accumulator.
+    pub fn is_parameterizable(&self) -> bool {
+        self.mode.is_some()
+    }
+
+    /// Get the field part (relative name or prefix) for a given field.
+    pub fn get_field_part(&self, field_name: &str) -> Option<&str> {
+        match &self.mode {
+            Some(PatternMode::Suffix { relatives }) => {
+                relatives.get(field_name).map(|s| s.as_str())
+            }
+            Some(PatternMode::Prefix { prefixes }) => prefixes.get(field_name).map(|s| s.as_str()),
+            Some(PatternMode::Templated { templates }) => {
+                templates.get(field_name).map(|s| s.as_str())
+            }
+            None => None,
+        }
+    }
+
+    /// Returns true if this pattern is in suffix mode.
+    pub fn is_suffix_mode(&self) -> bool {
+        matches!(
+            &self.mode,
+            Some(PatternMode::Suffix { .. } | PatternMode::Templated { .. })
+        )
+    }
+
+    /// Returns true if this pattern uses templated mode with a discriminator.
+    pub fn is_templated(&self) -> bool {
+        matches!(&self.mode, Some(PatternMode::Templated { .. }))
+    }
+
+    /// Extract the discriminator value from a concrete instance's field_parts.
+    /// Uses the pattern's templates to reverse-match and find the disc.
+    pub fn extract_disc_from_instance(
+        &self,
+        instance_field_parts: &BTreeMap<String, String>,
+    ) -> Option<String> {
+        let templates = match &self.mode {
+            Some(PatternMode::Templated { templates }) => templates,
+            _ => return None,
+        };
+        // Find a template with {disc} and extract the disc from the instance value.
+        // Strip leading underscore since _m() handles separators.
+        for (field_name, template) in templates {
+            if let Some(value) = instance_field_parts.get(field_name)
+                && let Some(disc) = extract_disc(template, value)
+            {
+                return Some(disc.trim_start_matches('_').to_string());
+            }
+        }
+        // If no template matched (all empty templates), disc is empty
+        Some(String::new())
+    }
+
+    /// Check if the given instance field parts match this pattern's field parts.
+    pub fn field_parts_match(&self, instance_field_parts: &BTreeMap<String, String>) -> bool {
+        match &self.mode {
+            Some(PatternMode::Suffix { relatives }) => {
+                relatives.iter().all(|(field_name, pattern_suffix)| {
+                    instance_field_parts
+                        .get(field_name)
+                        .is_some_and(|instance_suffix| instance_suffix == pattern_suffix)
+                })
+            }
+            Some(PatternMode::Prefix { prefixes }) => {
+                prefixes.iter().all(|(field_name, pattern_prefix)| {
+                    instance_field_parts
+                        .get(field_name)
+                        .is_some_and(|instance_prefix| instance_prefix == pattern_prefix)
+                })
+            }
+            Some(PatternMode::Templated { templates }) => {
+                // For templated patterns, check if the instance's field_parts
+                // can be produced by substituting some discriminator into the templates
+                let first_template_field = templates.iter().next();
+                let Some((ref_field, ref_template)) = first_template_field else {
+                    return false;
+                };
+                let Some(ref_value) = instance_field_parts.get(ref_field) else {
+                    return false;
+                };
+                // Extract discriminator from the reference field
+                let Some(disc) = extract_disc(ref_template, ref_value) else {
+                    return false;
+                };
+                // Verify all fields match with this discriminator
+                templates.iter().all(|(field_name, template)| {
+                    instance_field_parts
+                        .get(field_name)
+                        .is_some_and(|value| *value == template.replace("{disc}", &disc))
+                })
+            }
+            None => false,
+        }
+    }
+}
+
+/// Extract the discriminator value by matching a template against a concrete string.
+/// E.g., template `"ratio_{disc}_bps"` matched against `"ratio_pct99_bps"` yields `"pct99"`.
+fn extract_disc(template: &str, value: &str) -> Option<String> {
+    let (prefix, suffix) = template.split_once("{disc}")?;
+    if value.starts_with(prefix) && value.ends_with(suffix) {
+        let disc = &value[prefix.len()..value.len() - suffix.len()];
+        if !disc.is_empty() {
+            return Some(disc.to_string());
+        }
+    }
+    None
+}
+
+/// A field in a structural pattern.
+#[derive(Debug, Clone, PartialOrd, Ord)]
+pub struct PatternField {
+    /// Field name
+    pub name: String,
+    /// Rust type for leaves or pattern name for branches
+    pub rust_type: String,
+    /// JSON type from schema
+    pub json_type: String,
+    /// For leaves: the set of supported indexes. Empty for branches.
+    pub indexes: BTreeSet<Index>,
+    /// For branches referencing generic patterns: the concrete type parameter
+    pub type_param: Option<String>,
+}
+
+impl PatternField {
+    /// Returns true if this is a leaf field (has indexes).
+    pub fn is_leaf(&self) -> bool {
+        !self.indexes.is_empty()
+    }
+
+    /// Returns true if this is a branch field (no indexes).
+    pub fn is_branch(&self) -> bool {
+        self.indexes.is_empty()
+    }
+}
+
+impl std::hash::Hash for PatternField {
+    fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
+        self.name.hash(state);
+        self.rust_type.hash(state);
+        self.json_type.hash(state);
+        self.indexes.hash(state);
+    }
+}
+
+impl PartialEq for PatternField {
+    fn eq(&self, other: &Self) -> bool {
+        self.name == other.name
+            && self.rust_type == other.rust_type
+            && self.json_type == other.json_type
+            && self.indexes == other.indexes
+    }
+}
+
+impl Eq for PatternField {}

crates/brk_bindgen/tests/catalog_test.rs

@@ -0,0 +1,917 @@
+// //! Tests that verify pattern analysis using the real catalog.
+
+// use std::collections::{BTreeMap, BTreeSet};
+// use std::fmt::Write;
+
+// use brk_bindgen::ClientMetadata;
+// use brk_types::TreeNode;
+
+// /// Load the catalog from the JSON file.
+// fn load_catalog() -> TreeNode {
+//     let path = concat!(env!("CARGO_MANIFEST_DIR"), "/catalog.json");
+//     let catalog_json = std::fs::read_to_string(path).expect("Failed to read catalog.json");
+//     serde_json::from_str(&catalog_json).expect("Failed to parse catalog.json")
+// }
+
+// /// Load OpenAPI spec from openapi.json.
+// fn load_openapi_json() -> String {
+//     let path = concat!(env!("CARGO_MANIFEST_DIR"), "/openapi.json");
+//     std::fs::read_to_string(path).expect("Failed to read openapi.json")
+// }
+
+// /// Load metadata from the catalog.
+// #[allow(unused)]
+// fn load_metadata() -> ClientMetadata {
+//     ClientMetadata::from_catalog(load_catalog())
+// }
+
+// /// Collect all leaf metric names from a tree.
+// fn collect_leaf_names(node: &TreeNode, names: &mut BTreeSet<String>) {
+//     match node {
+//         TreeNode::Leaf(leaf) => {
+//             names.insert(leaf.name().to_string());
+//         }
+//         TreeNode::Branch(children) => {
+//             for child in children.values() {
+//                 collect_leaf_names(child, names);
+//             }
+//         }
+//     }
+// }
+
+// #[test]
+// fn test_catalog_loads() {
+//     let catalog = load_catalog();
+
+//     // Should be a branch with top-level categories
+//     let TreeNode::Branch(categories) = &catalog else {
+//         panic!("Expected catalog to be a branch");
+//     };
+
+//     // Check some expected top-level categories exist
+//     assert!(
+//         categories.contains_key("addresses"),
+//         "Missing addresses category"
+//     );
+//     assert!(categories.contains_key("blocks"), "Missing blocks category");
+//     assert!(categories.contains_key("market"), "Missing market category");
+//     assert!(categories.contains_key("supply"), "Missing supply category");
+
+//     println!("Catalog has {} top-level categories", categories.len());
+// }
+
+// #[test]
+// fn test_all_leaves_have_names() {
+//     let catalog = load_catalog();
+//     let mut names = BTreeSet::new();
+//     collect_leaf_names(&catalog, &mut names);
+
+//     println!("Catalog has {} unique metric names", names.len());
+//     assert!(!names.is_empty(), "Should have at least some metrics");
+
+//     // All names should be non-empty
+//     for name in &names {
+//         assert!(!name.is_empty(), "Found empty metric name");
+//     }
+// }
+
+// #[test]
+// fn test_pattern_detection() {
+//     let catalog = load_catalog();
+
+//     let (patterns, concrete_to_pattern, concrete_to_type_param, _node_bases) =
+//         brk_bindgen::detect_structural_patterns(&catalog);
+
+//     println!("Detected {} structural patterns", patterns.len());
+//     println!(
+//         "Concrete to pattern mappings: {}",
+//         concrete_to_pattern.len()
+//     );
+//     println!("Type parameter mappings: {}", concrete_to_type_param.len());
+
+//     // Print pattern details
+//     for pattern in &patterns {
+//         let mode_str = match &pattern.mode {
+//             Some(brk_bindgen::PatternMode::Suffix { relatives }) => {
+//                 format!("Suffix({})", relatives.len())
+//             }
+//             Some(brk_bindgen::PatternMode::Prefix { prefixes }) => {
+//                 format!("Prefix({})", prefixes.len())
+//             }
+//             None => "None".to_string(),
+//         };
+//         println!(
+//             "  {} (fields: {}, generic: {}, mode: {})",
+//             pattern.name,
+//             pattern.fields.len(),
+//             pattern.is_generic,
+//             mode_str
+//         );
+//     }
+
+//     // Should have detected some patterns
+//     assert!(!patterns.is_empty(), "Should detect at least some patterns");
+
+//     // Check that parameterizable patterns have valid modes
+//     for pattern in &patterns {
+//         if pattern.is_parameterizable() {
+//             let mode = pattern.mode.as_ref().unwrap();
+//             match mode {
+//                 brk_bindgen::PatternMode::Suffix { relatives } => {
+//                     assert_eq!(
+//                         relatives.len(),
+//                         pattern.fields.len(),
+//                         "Pattern {} should have relative for each field",
+//                         pattern.name
+//                     );
+//                 }
+//                 brk_bindgen::PatternMode::Prefix { prefixes } => {
+//                     assert_eq!(
+//                         prefixes.len(),
+//                         pattern.fields.len(),
+//                         "Pattern {} should have prefix for each field",
+//                         pattern.name
+//                     );
+//                 }
+//             }
+//         }
+//     }
+// }
+
+// #[test]
+// fn test_cost_basis_pattern() {
+//     let catalog = load_catalog();
+
+//     let (patterns, _, _, _) = brk_bindgen::detect_structural_patterns(&catalog);
+
+//     // Find CostBasisPattern2 and inspect it
+//     let cost_basis = patterns
+//         .iter()
+//         .find(|p| p.name == "CostBasisPattern2")
+//         .expect("CostBasisPattern2 should exist");
+
+//     println!("CostBasisPattern2:");
+//     println!(
+//         "  Fields: {:?}",
+//         cost_basis
+//             .fields
+//             .iter()
+//             .map(|f| &f.name)
+//             .collect::<Vec<_>>()
+//     );
+//     println!("  Mode: {:?}", cost_basis.mode);
+//     println!("  Is generic: {}", cost_basis.is_generic);
+
+//     // With suffix naming convention (cost_basis_max, cost_basis_min, cost_basis):
+//     //
+//     // At root level: common prefix is "cost_basis_" -> suffix mode
+//     //   max -> "max"
+//     //   min -> "min"
+//     //   percentiles -> "" (identity)
+//     //
+//     // At lth_ level: common prefix is "lth_cost_basis_" -> suffix mode
+//     //   max -> "max"
+//     //   min -> "min"
+//     //   percentiles -> "" (identity)
+//     //
+//     // Both use suffix mode with same relatives, so pattern IS parameterizable!
+//     assert!(
+//         cost_basis.is_parameterizable(),
+//         "CostBasisPattern2 should be parameterizable with consistent suffix mode"
+//     );
+// }
+
+// #[test]
+// fn test_realized_pattern3_fields() {
+//     let catalog = load_catalog();
+//     let metadata = ClientMetadata::from_catalog(catalog);
+
+//     let pattern = metadata
+//         .find_pattern("RealizedPattern3")
+//         .expect("RealizedPattern3 should exist");
+
+//     println!("RealizedPattern3 fields:");
+//     for field in &pattern.fields {
+//         let is_branch = field.is_branch();
+//         let is_pattern = metadata.find_pattern(&field.rust_type).is_some();
+//         let is_param = metadata.is_parameterizable(&field.rust_type);
+//         println!(
+//             "  {} -> {} (branch={}, pattern={}, param={})",
+//             field.name, field.rust_type, is_branch, is_pattern, is_param
+//         );
+//     }
+
+//     // Check if RealizedPattern3 is considered parameterizable
+//     println!(
+//         "\nRealizedPattern3 is_parameterizable (metadata): {}",
+//         metadata.is_parameterizable("RealizedPattern3")
+//     );
+// }
+
+// #[test]
+// fn test_parameterizable_patterns_have_mode() {
+//     let catalog = load_catalog();
+//     let (patterns, _, _, _) = brk_bindgen::detect_structural_patterns(&catalog);
+
+//     // All patterns that appear 2+ times should either:
+//     // 1. Be parameterizable (have a mode)
+//     // 2. Or have inconsistent instances (mode = None)
+//     //
+//     // Patterns with mode = None should be inlined, not generate factories
+
+//     let parameterizable: Vec<_> = patterns.iter().filter(|p| p.is_parameterizable()).collect();
+//     let non_parameterizable: Vec<_> = patterns
+//         .iter()
+//         .filter(|p| !p.is_parameterizable())
+//         .collect();
+
+//     println!("\nParameterizable patterns ({}):", parameterizable.len());
+//     for p in &parameterizable {
+//         let mode = p.mode.as_ref().unwrap();
+//         let mode_type = match mode {
+//             brk_bindgen::PatternMode::Suffix { .. } => "Suffix",
+//             brk_bindgen::PatternMode::Prefix { .. } => "Prefix",
+//         };
+//         println!("  {} ({} fields, {})", p.name, p.fields.len(), mode_type);
+//     }
+
+//     println!(
+//         "\nNon-parameterizable patterns ({}):",
+//         non_parameterizable.len()
+//     );
+//     for p in &non_parameterizable {
+//         println!("  {} ({} fields)", p.name, p.fields.len());
+//     }
+
+//     // Verify all parameterizable patterns have valid modes with all fields
+//     for pattern in &parameterizable {
+//         let mode = pattern.mode.as_ref().unwrap();
+//         let field_names: BTreeSet<_> = pattern.fields.iter().map(|f| f.name.clone()).collect();
+
+//         match mode {
+//             brk_bindgen::PatternMode::Suffix { relatives } => {
+//                 let mode_fields: BTreeSet<_> = relatives.keys().cloned().collect();
+//                 assert_eq!(
+//                     field_names, mode_fields,
+//                     "Pattern {} suffix mode should have all fields",
+//                     pattern.name
+//                 );
+//             }
+//             brk_bindgen::PatternMode::Prefix { prefixes } => {
+//                 let mode_fields: BTreeSet<_> = prefixes.keys().cloned().collect();
+//                 assert_eq!(
+//                     field_names, mode_fields,
+//                     "Pattern {} prefix mode should have all fields",
+//                     pattern.name
+//                 );
+//             }
+//         }
+//     }
+// }
+
+// #[test]
+// fn test_fee_rate_pattern_relatives() {
+//     let catalog = load_catalog();
+//     let (patterns, _, _, _) = brk_bindgen::detect_structural_patterns(&catalog);
+
+//     let fee_rate_pattern = patterns
+//         .iter()
+//         .find(|p| p.name == "FeeRatePattern")
+//         .expect("FeeRatePattern should exist");
+
+//     println!("FeeRatePattern mode:");
+//     if let Some(mode) = &fee_rate_pattern.mode {
+//         match mode {
+//             brk_bindgen::PatternMode::Suffix { relatives } => {
+//                 println!("  Suffix mode:");
+//                 for (field, relative) in relatives {
+//                     println!("    {} -> '{}'", field, relative);
+//                 }
+//             }
+//             brk_bindgen::PatternMode::Prefix { prefixes } => {
+//                 println!("  Prefix mode:");
+//                 for (field, prefix) in prefixes {
+//                     println!("    {} -> '{}'", field, prefix);
+//                 }
+//             }
+//         }
+//     } else {
+//         println!("  No mode (not parameterizable)");
+//     }
+
+//     // Check that relatives are correct - should be "average", "max", etc.
+//     // NOT "tx_weight_average", "tx_weight_max", etc.
+//     if let Some(brk_bindgen::PatternMode::Suffix { relatives }) = &fee_rate_pattern.mode {
+//         assert_eq!(
+//             relatives.get("average"),
+//             Some(&"average".to_string()),
+//             "average relative should be 'average', not 'tx_weight_average'"
+//         );
+//     }
+// }
+
+// #[test]
+// fn test_index_patterns() {
+//     let catalog = load_catalog();
+
+//     let index_patterns = brk_bindgen::detect_index_patterns(&catalog);
+
+//     // println!("Used indexes: {:?}", used_indexes);
+//     println!("Index set patterns: {}", index_patterns.len());
+
+//     for pattern in &index_patterns {
+//         println!("  {} -> {:?}", pattern.name, pattern.indexes);
+//     }
+
+//     // Should have detected some index patterns
+//     assert!(!index_patterns.is_empty(), "Should detect index patterns");
+// }
+
+// #[test]
+// fn test_generated_rust_output() {
+//     let catalog = load_catalog();
+//     let metadata = ClientMetadata::from_catalog(catalog.clone());
+
+//     // Collect all metric names from the catalog
+//     let mut all_metrics = BTreeSet::new();
+//     collect_leaf_names(&catalog, &mut all_metrics);
+
+//     // Generate Rust client output
+//     let mut rust_output = String::new();
+//     brk_bindgen::rust::client::generate_imports(&mut rust_output);
+//     brk_bindgen::rust::client::generate_base_client(&mut rust_output);
+//     brk_bindgen::rust::client::generate_metric_pattern_trait(&mut rust_output);
+//     brk_bindgen::rust::client::generate_endpoint(&mut rust_output);
+//     brk_bindgen::rust::client::generate_index_accessors(
+//         &mut rust_output,
+//         &metadata.index_set_patterns,
+//     );
+//     brk_bindgen::rust::client::generate_pattern_structs(
+//         &mut rust_output,
+//         &metadata.structural_patterns,
+//         &metadata,
+//     );
+//     brk_bindgen::rust::tree::generate_tree(&mut rust_output, &metadata.catalog, &metadata);
+//     brk_bindgen::rust::api::generate_main_client(&mut rust_output, &[]);
+
+//     // Count metrics that appear as direct string literals
+//     let mut direct_metrics = 0;
+//     for metric in &all_metrics {
+//         if rust_output.contains(&format!("\"{}\"", metric)) {
+//             direct_metrics += 1;
+//         }
+//     }
+
+//     println!("\nGenerated Rust output stats:");
+//     println!("  Total metrics in catalog: {}", all_metrics.len());
+//     println!("  Direct string literals: {}", direct_metrics);
+//     println!(
+//         "  Via pattern factories: {}",
+//         all_metrics.len() - direct_metrics
+//     );
+//     println!("  Output size: {} bytes", rust_output.len());
+
+//     // Write output to test directory (not actual client)
+//     let output_dir = concat!(env!("CARGO_MANIFEST_DIR"), "/tests/output");
+//     std::fs::create_dir_all(output_dir).ok();
+//     let output_path = format!("{}/rust_client.rs", output_dir);
+//     std::fs::write(&output_path, &rust_output).expect("Failed to write client output");
+//     println!("  Wrote output to: {}", output_path);
+
+//     // Verify the output contains the key components
+//     assert!(rust_output.contains("fn _m("), "Should define _m helper");
+//     assert!(
+//         rust_output.contains("pub struct MetricsTree"),
+//         "Should have MetricsTree"
+//     );
+//     assert!(
+//         rust_output.contains("impl MetricsTree"),
+//         "Should have MetricsTree impl"
+//     );
+
+//     // Count parameterizable patterns (these use _m for dynamic metric names)
+//     // Use metadata.is_parameterizable() for full recursive check
+//     let parameterizable_count = metadata
+//         .structural_patterns
+//         .iter()
+//         .filter(|p| metadata.is_parameterizable(&p.name))
+//         .count();
+//     println!("  Parameterizable patterns: {}", parameterizable_count);
+
+//     // Verify all pattern structs are generated (parameterizable and non)
+//     for pattern in &metadata.structural_patterns {
+//         assert!(
+//             rust_output.contains(&format!("pub struct {}", pattern.name)),
+//             "Missing pattern struct: {}",
+//             pattern.name
+//         );
+//     }
+
+//     println!("\nGenerated Rust client is complete!");
+// }
+
+// #[test]
+// fn test_generated_javascript_output() {
+//     let catalog = load_catalog();
+//     let metadata = ClientMetadata::from_catalog(catalog.clone());
+
+//     // Collect all metric names from the catalog
+//     let mut all_metrics = BTreeSet::new();
+//     collect_leaf_names(&catalog, &mut all_metrics);
+
+//     // Load schemas from OpenAPI spec only (catalog schemas require runtime data)
+//     let openapi_json = load_openapi_json();
+//     let schemas = brk_bindgen::extract_schemas(&openapi_json);
+
+//     // Generate JavaScript client output
+//     let mut js_output = String::new();
+//     writeln!(js_output, "// Auto-generated BRK JavaScript client").unwrap();
+//     writeln!(js_output, "// Do not edit manually\n").unwrap();
+//     brk_bindgen::javascript::types::generate_type_definitions(&mut js_output, &schemas);
+//     brk_bindgen::javascript::client::generate_base_client(&mut js_output);
+//     brk_bindgen::javascript::client::generate_index_accessors(
+//         &mut js_output,
+//         &metadata.index_set_patterns,
+//     );
+//     brk_bindgen::javascript::client::generate_structural_patterns(
+//         &mut js_output,
+//         &metadata.structural_patterns,
+//         &metadata,
+//     );
+//     brk_bindgen::javascript::tree::generate_tree_typedefs(
+//         &mut js_output,
+//         &metadata.catalog,
+//         &metadata,
+//     );
+//     brk_bindgen::javascript::tree::generate_main_client(
+//         &mut js_output,
+//         &metadata.catalog,
+//         &metadata,
+//         &[],
+//     );
+
+//     // Count metrics that appear as direct string literals
+//     let mut direct_metrics = 0;
+//     for metric in &all_metrics {
+//         if js_output.contains(&format!("'{}'", metric))
+//             || js_output.contains(&format!("\"{}\"", metric))
+//         {
+//             direct_metrics += 1;
+//         }
+//     }
+
+//     println!("\nGenerated JavaScript output stats:");
+//     println!("  Total metrics in catalog: {}", all_metrics.len());
+//     println!("  Direct string literals: {}", direct_metrics);
+//     println!(
+//         "  Via pattern factories: {}",
+//         all_metrics.len() - direct_metrics
+//     );
+//     println!("  Output size: {} bytes", js_output.len());
+//     println!("  Output lines: {}", js_output.lines().count());
+
+//     // Write output to test directory (not actual client)
+//     let output_dir = concat!(env!("CARGO_MANIFEST_DIR"), "/tests/output");
+//     std::fs::create_dir_all(output_dir).ok();
+//     let output_path = format!("{}/js_client.js", output_dir);
+//     std::fs::write(&output_path, &js_output).expect("Failed to write JS client output");
+//     println!("  Wrote output to: {}", output_path);
+
+//     // Verify the output contains key components
+//     assert!(js_output.contains("const _m ="), "Should define _m helper");
+//     assert!(js_output.contains("const _p ="), "Should define _p helper");
+//     assert!(
+//         js_output.contains("@typedef {Object} MetricsTree"),
+//         "Should have MetricsTree typedef"
+//     );
+//     assert!(
+//         js_output.contains("class BrkClient"),
+//         "Should have BrkClient class"
+//     );
+
+//     // Verify all pattern factories are generated
+//     for pattern in &metadata.structural_patterns {
+//         assert!(
+//             js_output.contains(&format!("function create{}(", pattern.name)),
+//             "Missing pattern factory: {}",
+//             pattern.name
+//         );
+//     }
+
+//     println!("\nGenerated JavaScript client is complete!");
+// }
+
+// #[test]
+// fn test_generated_python_output() {
+//     let catalog = load_catalog();
+//     let metadata = ClientMetadata::from_catalog(catalog.clone());
+
+//     // Collect all metric names from the catalog
+//     let mut all_metrics = BTreeSet::new();
+//     collect_leaf_names(&catalog, &mut all_metrics);
+
+//     // Load schemas from OpenAPI spec only (catalog schemas require runtime data)
+//     let openapi_json = load_openapi_json();
+//     let schemas = brk_bindgen::extract_schemas(&openapi_json);
+
+//     // Generate Python client output
+//     let mut py_output = String::new();
+//     writeln!(py_output, "# Auto-generated BRK Python client").unwrap();
+//     writeln!(py_output, "# Do not edit manually\n").unwrap();
+//     writeln!(py_output, "from typing import TypeVar, Generic, Any, Optional, List, Literal, TypedDict, Union, Protocol, overload, Iterator, Tuple, TYPE_CHECKING").unwrap();
+//     writeln!(py_output, "\nif TYPE_CHECKING:").unwrap();
+//     writeln!(py_output, "    import pandas as pd  # type: ignore[import-not-found]").unwrap();
+//     writeln!(py_output, "    import polars as pl  # type: ignore[import-not-found]").unwrap();
+//     writeln!(
+//         py_output,
+//         "from http.client import HTTPSConnection, HTTPConnection"
+//     )
+//     .unwrap();
+//     writeln!(py_output, "from urllib.parse import urlparse").unwrap();
+//     writeln!(py_output, "from datetime import date, timedelta").unwrap();
+//     writeln!(py_output, "from dataclasses import dataclass").unwrap();
+//     writeln!(py_output, "import json\n").unwrap();
+//     writeln!(py_output, "T = TypeVar('T')\n").unwrap();
+
+//     brk_bindgen::python::types::generate_type_definitions(&mut py_output, &schemas);
+//     brk_bindgen::python::client::generate_base_client(&mut py_output);
+//     brk_bindgen::python::client::generate_endpoint_class(&mut py_output);
+//     brk_bindgen::python::client::generate_index_accessors(
+//         &mut py_output,
+//         &metadata.index_set_patterns,
+//     );
+//     brk_bindgen::python::client::generate_structural_patterns(
+//         &mut py_output,
+//         &metadata.structural_patterns,
+//         &metadata,
+//     );
+//     brk_bindgen::python::tree::generate_tree_classes(&mut py_output, &metadata.catalog, &metadata);
+//     brk_bindgen::python::api::generate_main_client(&mut py_output, &[]);
+
+//     // Count metrics that appear as direct string literals
+//     let mut direct_metrics = 0;
+//     for metric in &all_metrics {
+//         if py_output.contains(&format!("'{}'", metric))
+//             || py_output.contains(&format!("\"{}\"", metric))
+//         {
+//             direct_metrics += 1;
+//         }
+//     }
+
+//     println!("\nGenerated Python output stats:");
+//     println!("  Total metrics in catalog: {}", all_metrics.len());
+//     println!("  Direct string literals: {}", direct_metrics);
+//     println!(
+//         "  Via pattern factories: {}",
+//         all_metrics.len() - direct_metrics
+//     );
+//     println!("  Output size: {} bytes", py_output.len());
+//     println!("  Output lines: {}", py_output.lines().count());
+
+//     // Write output to test directory (not actual client)
+//     let output_dir = concat!(env!("CARGO_MANIFEST_DIR"), "/tests/output");
+//     std::fs::create_dir_all(output_dir).ok();
+//     let output_path = format!("{}/python_client.py", output_dir);
+//     std::fs::write(&output_path, &py_output).expect("Failed to write Python client output");
+//     println!("  Wrote output to: {}", output_path);
+
+//     // Verify the output contains key components
+//     assert!(py_output.contains("def _m("), "Should define _m helper");
+//     assert!(py_output.contains("def _p("), "Should define _p helper");
+//     assert!(
+//         py_output.contains("class MetricsTree:"),
+//         "Should have MetricsTree class"
+//     );
+//     assert!(
+//         py_output.contains("class BrkClient"),
+//         "Should have BrkClient class"
+//     );
+
+//     // Verify all pattern classes have constructors
+//     for pattern in &metadata.structural_patterns {
+//         assert!(
+//             py_output.contains(&format!("class {}:", pattern.name))
+//                 || py_output.contains(&format!("class {}(", pattern.name)),
+//             "Missing pattern class: {}",
+//             pattern.name
+//         );
+//     }
+
+//     println!("\nGenerated Python client is complete!");
+// }
+
+// #[test]
+// fn test_cost_basis_relatives() {
+//     let catalog = load_catalog();
+
+//     // Find cost_basis branches that have 3 direct children (max, min, percentiles)
+//     fn find_cost_basis_with_percentiles(
+//         node: &TreeNode,
+//         path: &str,
+//     ) -> Vec<(String, Vec<(String, String)>)> {
+//         let mut results = Vec::new();
+//         if let TreeNode::Branch(children) = node {
+//             for (name, child) in children {
+//                 let child_path = if path.is_empty() {
+//                     name.clone()
+//                 } else {
+//                     format!("{}.{}", path, name)
+//                 };
+
+//                 if name == "cost_basis"
+//                     && let TreeNode::Branch(cb_children) = child
+//                     && cb_children.contains_key("percentiles")
+//                 {
+//                     // Found a cost_basis with percentiles
+//                     let mut metrics = Vec::new();
+//                     for (field_name, field_node) in cb_children {
+//                         match field_node {
+//                             TreeNode::Leaf(leaf) => {
+//                                 metrics.push((field_name.clone(), leaf.name().to_string()));
+//                             }
+//                             TreeNode::Branch(pct_children) => {
+//                                 // Get first percentile as example
+//                                 if let Some((_, TreeNode::Leaf(first))) = pct_children.iter().next()
+//                                 {
+//                                     metrics.push((
+//                                         format!("{}.first", field_name),
+//                                         first.name().to_string(),
+//                                     ));
+//                                 }
+//                             }
+//                         }
+//                     }
+//                     results.push((child_path.clone(), metrics));
+//                 }
+//                 results.extend(find_cost_basis_with_percentiles(child, &child_path));
+//             }
+//         }
+//         results
+//     }
+
+//     let instances = find_cost_basis_with_percentiles(&catalog, "");
+
+//     println!("\nCostBasisPattern2 instances (with percentiles):");
+//     for (path, metrics) in instances.iter().take(10) {
+//         println!("  {}:", path);
+//         for (field, metric) in metrics {
+//             println!("    {} -> {}", field, metric);
+//         }
+//     }
+
+//     // Now compute what relatives the pattern detection would see
+//     // The key is: percentiles returns its BASE (common prefix of pct05, pct10, etc.)
+//     // not the individual percentile metrics
+//     use brk_bindgen::find_common_prefix;
+
+//     println!("\nComputing relatives (simulating branch base returns):");
+//     for (path, metrics) in instances.iter().take(5) {
+//         println!("  Instance: {}", path);
+
+//         // For leaves (max, min), the base is the metric name
+//         // For branches (percentiles), the base is the common prefix of its children
+//         let mut child_bases: std::collections::BTreeMap<String, String> =
+//             std::collections::BTreeMap::new();
+//         for (field, metric) in metrics {
+//             if field.starts_with("percentiles.") {
+//                 // This is a percentile metric - compute what the percentiles branch would return
+//                 // The base is the metric name with the pct suffix stripped
+//                 let base = metric
+//                     .strip_suffix("_pct05")
+//                     .or_else(|| metric.strip_suffix("_pct10"))
+//                     .unwrap_or(metric)
+//                     .to_string();
+//                 child_bases.insert("percentiles".to_string(), base);
+//             } else {
+//                 child_bases.insert(field.clone(), metric.clone());
+//             }
+//         }
+
+//         let bases: Vec<&str> = child_bases.values().map(|s| s.as_str()).collect();
+//         println!("    Child bases:");
+//         for (field, base) in &child_bases {
+//             println!("      {} -> {}", field, base);
+//         }
+
+//         if let Some(prefix) = find_common_prefix(&bases) {
+//             println!("    Common prefix: '{}'", prefix);
+//             for (field, base) in &child_bases {
+//                 let relative = base.strip_prefix(&prefix).unwrap_or(base);
+//                 println!("      {} -> relative '{}'", field, relative);
+//             }
+//         } else {
+//             println!("    No common prefix found!");
+//         }
+//     }
+// }
+
+// #[test]
+// fn test_debug_cost_basis_pattern2_mode() {
+//     // Debug why CostBasisPattern2 has mode=None
+//     let catalog = load_catalog();
+//     let metadata = brk_bindgen::ClientMetadata::from_catalog(catalog.clone());
+//     let pattern_lookup = metadata.pattern_lookup();
+
+//     let pattern = metadata
+//         .find_pattern("CostBasisPattern2")
+//         .expect("CostBasisPattern2 should exist");
+
+//     println!("\nCostBasisPattern2 fields:");
+//     for field in &pattern.fields {
+//         println!("  {} (type: {})", field.name, field.rust_type);
+//     }
+//     println!("Mode: {:?}", pattern.mode);
+
+//     // Now debug the instance collection
+//     #[derive(Debug, Clone)]
+//     struct DebugInstanceAnalysis {
+//         base: String,
+//         field_parts: std::collections::BTreeMap<String, String>,
+//         is_suffix_mode: bool,
+//     }
+
+//     fn collect_debug(
+//         node: &TreeNode,
+//         pattern_lookup: &std::collections::BTreeMap<Vec<brk_bindgen::PatternField>, String>,
+//         all_analyses: &mut std::collections::BTreeMap<String, Vec<DebugInstanceAnalysis>>,
+//     ) -> Option<String> {
+//         match node {
+//             TreeNode::Leaf(leaf) => Some(leaf.name().to_string()),
+//             TreeNode::Branch(children) => {
+//                 let mut child_bases: std::collections::BTreeMap<String, String> =
+//                     std::collections::BTreeMap::new();
+//                 for (field_name, child_node) in children {
+//                     if let Some(base) = collect_debug(child_node, pattern_lookup, all_analyses) {
+//                         child_bases.insert(field_name.clone(), base);
+//                     }
+//                 }
+
+//                 if child_bases.is_empty() {
+//                     return None;
+//                 }
+
+//                 // Analyze this instance
+//                 let bases: Vec<&str> = child_bases.values().map(|s| s.as_str()).collect();
+//                 let (base, field_parts, is_suffix_mode) =
+//                     if let Some(common_prefix) = brk_bindgen::find_common_prefix(&bases) {
+//                         let base = common_prefix.trim_end_matches('_').to_string();
+//                         let mut parts = std::collections::BTreeMap::new();
+//                         for (field_name, child_base) in &child_bases {
+//                             let relative = if *child_base == base {
+//                                 String::new()
+//                             } else {
+//                                 child_base
+//                                     .strip_prefix(&common_prefix)
+//                                     .unwrap_or(child_base)
+//                                     .to_string()
+//                             };
+//                             parts.insert(field_name.clone(), relative);
+//                         }
+//                         (base, parts, true)
+//                     } else {
+//                         let base = child_bases.values().next().cloned().unwrap_or_default();
+//                         let parts = child_bases
+//                             .iter()
+//                             .map(|(k, v)| (k.clone(), v.clone()))
+//                             .collect();
+//                         (base, parts, true)
+//                     };
+
+//                 let analysis = DebugInstanceAnalysis {
+//                     base: base.clone(),
+//                     field_parts,
+//                     is_suffix_mode,
+//                 };
+
+//                 // Get the pattern name for this node
+//                 let fields = brk_bindgen::get_node_fields(children, pattern_lookup);
+//                 if let Some(pattern_name) = pattern_lookup.get(&fields) {
+//                     all_analyses
+//                         .entry(pattern_name.clone())
+//                         .or_default()
+//                         .push(analysis);
+//                 }
+
+//                 Some(base)
+//             }
+//         }
+//     }
+
+//     let mut all_analyses: BTreeMap<String, Vec<DebugInstanceAnalysis>> = BTreeMap::new();
+//     collect_debug(&catalog, &pattern_lookup, &mut all_analyses);
+
+//     if let Some(analyses) = all_analyses.get("CostBasisPattern2") {
+//         println!(
+//             "\nCollected {} instances of CostBasisPattern2:",
+//             analyses.len()
+//         );
+//         for (i, a) in analyses.iter().enumerate() {
+//             println!("  Instance {}:", i);
+//             println!("    base: {}", a.base);
+//             println!("    is_suffix: {}", a.is_suffix_mode);
+//             println!("    field_parts:");
+//             for (f, p) in &a.field_parts {
+//                 println!("      {} -> '{}'", f, p);
+//             }
+//         }
+
+//         // Check consistency
+//         if analyses.len() >= 2 {
+//             let first = &analyses[0];
+//             for (i, a) in analyses.iter().enumerate().skip(1) {
+//                 if a.is_suffix_mode != first.is_suffix_mode {
+//                     println!("  INCONSISTENT: Instance {} has different mode", i);
+//                 }
+//                 for (field, part) in &a.field_parts {
+//                     if first.field_parts.get(field) != Some(part) {
+//                         println!(
+//                             "  INCONSISTENT: Instance {} field '{}' has part '{}' vs '{}'",
+//                             i,
+//                             field,
+//                             part,
+//                             first
+//                                 .field_parts
+//                                 .get(field)
+//                                 .unwrap_or(&"<missing>".to_string())
+//                         );
+//                     }
+//                 }
+//             }
+//         }
+//     } else {
+//         println!("\nNo instances collected for CostBasisPattern2!");
+//     }
+// }
+
+// #[test]
+// fn test_root_cost_basis_prefix() {
+//     use brk_bindgen::find_common_prefix;
+
+//     // Root-level cost_basis has:
+//     // max -> "max_cost_basis"
+//     // min -> "min_cost_basis"
+//     // percentiles -> "cost_basis" (base of pct05, pct10, etc.)
+
+//     let bases = vec!["max_cost_basis", "min_cost_basis", "cost_basis"];
+//     let prefix = find_common_prefix(&bases);
+//     println!("Root cost_basis prefix: {:?}", prefix);
+
+//     // Compare with nested cost_basis
+//     let nested_bases = vec![
+//         "utxos_at_least_15y_old_max_cost_basis",
+//         "utxos_at_least_15y_old_min_cost_basis",
+//         "utxos_at_least_15y_old_cost_basis",
+//     ];
+//     let nested_prefix = find_common_prefix(&nested_bases);
+//     println!("Nested cost_basis prefix: {:?}", nested_prefix);
+// }
+
+// #[test]
+// fn test_utxo_cohorts_all_activity_base() {
+//     // Test that distribution.utxo_cohorts.all.activity uses empty base
+//     // because its children (coinblocks_destroyed, coindays_destroyed, etc.)
+//     // have no common prefix or suffix.
+//     let catalog = load_catalog();
+//     let metadata = ClientMetadata::from_catalog(catalog.clone());
+
+//     // Generate JavaScript output
+//     let mut js_output = String::new();
+//     writeln!(js_output, "// Test output").unwrap();
+//     brk_bindgen::javascript::client::generate_base_client(&mut js_output);
+//     brk_bindgen::javascript::client::generate_index_accessors(
+//         &mut js_output,
+//         &metadata.index_set_patterns,
+//     );
+//     brk_bindgen::javascript::client::generate_structural_patterns(
+//         &mut js_output,
+//         &metadata.structural_patterns,
+//         &metadata,
+//     );
+//     brk_bindgen::javascript::tree::generate_tree_typedefs(
+//         &mut js_output,
+//         &metadata.catalog,
+//         &metadata,
+//     );
+//     brk_bindgen::javascript::tree::generate_main_client(
+//         &mut js_output,
+//         &metadata.catalog,
+//         &metadata,
+//         &[],
+//     );
+
+//     // The all.activity should use empty base, so metrics don't get duplicated
+//     // Look for: activity: createActivityPattern2(this, '')
+//     // NOT: activity: createActivityPattern2(this, 'coinblocks_destroyed')
+//     assert!(
+//         !js_output.contains("createActivityPattern2(this, 'coinblocks_destroyed')"),
+//         "all.activity should NOT use 'coinblocks_destroyed' as base (causes duplication)"
+//     );
+
+//     // Check that it uses empty string as base
+//     assert!(
+//         js_output.contains("activity: createActivityPattern2(this, '')"),
+//         "all.activity should use empty base"
+//     );
+
+//     println!("utxo_cohorts.all.activity base test passed!");
+// }

crates/brk_bundler/Cargo.toml

@@ -1,18 +0,0 @@
-[package]
-name = "brk_bundler"
-description = "A thin wrapper around rolldown"
-version.workspace = true
-edition.workspace = true
-license.workspace = true
-homepage.workspace = true
-repository.workspace = true
-rust-version.workspace = true
-build = "build.rs"
-
-[dependencies]
-log = { workspace = true }
-notify = "8.2.0"
-brk_rolldown = "0.1.5"
-# brk_rolldown = { path = "../../../rolldown/crates/rolldown"}
-sugar_path = "1.2.0"
-tokio = { workspace = true }