@gomani/bench

The proof-of-thesis harness: it measures the framework against its own claims with real numbers, and never fabricates — unmeasured metrics are labelled, not invented.

gomani bench                 # measure the targets in gomani.bench.json → a Markdown report
gomani bench --out RESULTS.md
gomani bench --runtime       # also run the browser probe (needs Playwright + a browser)

What it measures

• Shipped weight (always, no browser). For each built dist, the eager JS — the transitive closure of JavaScript reachable by static imports from the page's entry scripts (dynamic import() — Gomani's islands, or any code-split chunk — is counted as lazy, not eager) — plus HTML and CSS, all as Brotli-compressed wire bytes. Framework-agnostic: it reads the dist of any tool, so Gomani and the bench-baselines/* (React SPA, etc.) are measured identically. This carries H1 (client cost).
• Offline shell / PWA (H4). Whether an app-shell service worker + PWA manifest are present.
• .goma vs JSX parity. A target may declare "compareTo": "<jsx-twin>"; the report then asserts that target's eager JS ≤ the twin's and prints a .goma-vs-JSX table. Since both surfaces lower to the same runtime, a .goma port must ship no more client JS than the hand-written JSX of the same app — this measures it (starter-gomani ships a 0 B delta vs starter).
• Runtime (optional, Playwright). DOM-interactive, LCP, JS heap, and actual transfer under throttled MBD/MBN profiles. Gated: absent a browser, these columns are —, and the static headline stands.

The report

A Markdown table of shipped weight per target, plus the H1–H9 hypothesis table — with H1 and H4 filled straight from the measurement and the rest labelled by how they are covered (a component's tests, a process claim, or on-device). See bench-results.md for a committed run.

Sample headline: for the same content, Gomani's griot ships 3.3 KB of eager JS; a React SPA ships 39.3 KB (Brotli).

Programmatic API: measureDist, runBench, formatBenchReport, measureRuntime, HYPOTHESES, PROFILES.