chore: update benchmarks for v0.2

.github/workflows/check.yml

@@ -7,7 +7,7 @@ on:
   push:
     branches: [main]
   pull_request:
-    branches: ['**']
+    branches: ["**"]
   # Allow manually re-running the check from the Actions tab.
   workflow_dispatch:
 
@@ -28,7 +28,7 @@ jobs:
 
     strategy:
       matrix:
-        node-version: ['24.14']
+        node-version: ["24.14"]
 
     steps:
       # persist-credentials: false keeps the GITHUB_TOKEN out of .git/config, so a

CLAUDE.md

@@ -23,19 +23,21 @@ IMPORTANT for AI agents: Do NOT run `gro dev` - the developer will manage the de
 
 - Svelte 5 - component framework with runes
 - SvelteKit - application framework with static adapter
-- fuz_css (`@fuzdev/fuz_css`) - CSS framework and design system
+- fuz_css (`@fuzdev/fuz_css`) - semantic-first CSS framework and design system
 - fuz_ui (`@fuzdev/fuz_ui`) - UI components, theming, docs system
 - fuz_util (`@fuzdev/fuz_util`) - utility functions
 - fuz_code (`@fuzdev/fuz_code`) - syntax highlighting
 - Gro (`@fuzdev/gro`) - build system and task runner
-- prettier + prettier-plugin-svelte - code formatting
+- tsv (via Gro) - code formatting
 - mdz (`@fuzdev/mdz`) - markdown preprocessor wired into `svelte.config.js`
 - `@fuzdev/tsv_wasm` - tsv's formatter + parser as WASM; powers the playground, loaded lazily in the browser
 
-Note: `@fuzdev/tsv_wasm` is loaded only on `/docs/playground` via a browser-only dynamic `import()`, so the ~900KB WASM never weighs down `/docs` or the prerendered pages.
+Note: `@fuzdev/tsv_wasm` is loaded only on `/docs/playground` via a browser-only dynamic `import()`, so the ~900KB-gzipped WASM (~2.9MB decoded) never weighs down `/docs` or the prerendered pages.
 
 Note: several devDependencies — `@webref/css` (CSS spec data), `zimmerframe` (AST traversal), `@sveltejs/acorn-typescript`, `zod`, and `@fuzdev/blake3_wasm` — are *optional peer dependencies* of `@fuzdev/fuz_css`'s `vite_plugin_fuz_css`, declared here so its build-time CSS generation resolves them (e.g. `css_literal.ts` imports `@webref/css`, `css_class_extractor.ts` walks ASTs with `zimmerframe`). They aren't imported by this app's own source, so don't mistake them for dead deps.
 
+Note: `vite` is deliberately held at 7.x (with `@sveltejs/vite-plugin-svelte` 6.x) — vite 8 + plugin-svelte 7 was buggy with this app or SvelteKit's integration. Don't upgrade to vite 8 without deliberately re-verifying the site works.
+
 ## Scope
 
 tsv.fuz.dev is the public face of the tsv tool:
@@ -56,8 +58,6 @@ tsv.fuz.dev is the public face of the tsv tool:
 
 ```
 src/
-├── lib/
-│   └── sample.ts             # placeholder constant (not yet wired up)
 ├── routes/
 │   ├── +page.svelte          # Home: hero, links to docs and benchmarks
 │   ├── +layout.svelte        # Root layout: fuz_css/fuz_code CSS, ThemeRoot, SiteState
@@ -81,23 +81,33 @@ src/
 Benchmark data comes from `tsv`. Full workflow to update:
 
 ```bash
-# 1. In ~/dev/tsv — run benchmarks (builds ffi + wasm automatically)
+# 1. In ~/dev/tsv — run benchmarks across deno/node/bun (builds artifacts automatically)
 deno task bench
 
-# 2. In ~/dev/tsv.fuz.dev — copy the latest result
+# 2. In ~/dev/tsv.fuz.dev — copy the latest results
 npm run update-benchmarks
 ```
 
-Step 1 writes `benches/deno/results/report.json` (committed to tsv).
-Step 2 copies it to `src/routes/docs/benchmarks/benchmarks.json`.
-The JSON format matches `BenchmarkBaseline` in `benchmark_data.ts`.
+Step 1 writes the per-runtime `benches/js/results/report.<runtime>.{json,md}`
+siblings, the composed cross-runtime `report.{json,md}`, and the conformance
+coverage report `report.conformance.node.json` (committed to tsv).
+Step 2 copies three of them — `report.node.json` → `benchmarks.json`, the
+composed `report.json` → `benchmarks_cross_runtime.json`, and
+`report.conformance.node.json` → `benchmarks_conformance.json` — then runs
+`gro format` over the copies. Note the script's source paths are hardcoded to
+`../tsv` — if the reports were generated in a different worktree, copy them
+into `~/dev/tsv` (or copy manually) first.
+The JSON formats match the types in `benchmark_data.ts`.
 
 Key files in `src/routes/docs/benchmarks/`:
 
-- `benchmarks.json` — raw benchmark data (copied from tsv)
+- `benchmarks.json` — per-runtime Node report (copied from tsv)
+- `benchmarks_cross_runtime.json` — composed cross-runtime report (copied from tsv)
+- `benchmarks_conformance.json` — conformance parse-coverage report (copied from tsv)
 - `benchmark_data.ts` — TypeScript types matching the JSON format
-- `benchmarks.ts` — re-exports the JSON with types
+- `benchmarks.ts`, `benchmarks_cross_runtime.ts`, `benchmarks_conformance.ts` — re-export the JSON with types
 - `BenchmarksBar.svelte`, `BenchmarksGroup.svelte`, etc. — visualization components
+- `BenchmarksBaselineGroup.svelte` — shared interactive column behind the format, parse, and binary-size groups: hovering a row re-baselines that group's ratios (each of the three groups per section is independent), restoring the default anchor (the canonical reference — Prettier for format, the JS baseline for parse — and the smallest build for size) on mouseleave. `benchmark_data.ts`'s `compute_baseline_ratio`/`format_baseline_ratio`/`baseline_ratio_color` carry the per-`BaselineDirection` (`speed`/`size`) formulas it and the derivations share
 
 ## Architecture
 
@@ -117,7 +127,7 @@ Deploys to `https://tsv.fuz.dev/` via `gro deploy` (builds and pushes to deploy
 
 - TypeScript strict mode
 - Svelte 5 with runes API
-- Prettier with tabs, 100 char width
+- tsv with tabs, 100 char width
 - Node >= 24.14
 - Private package (not published to npm)