perf(wasix): qualify browser benchmarks

[f0rr0]

Summary

Adds qualified browser performance evidence as a minimal follow-up to merged #133, stacked directly on the current #129 head.

• defines a checked-in direct/direct and worker/worker benchmark plan against exact @electric-sql/pglite@0.5.4
• uses the same comfortable-win statistic as the Node harness, but requires the <= 0.80 geometric-mean ratio independently in both browser topologies so one cannot subsidize the other
• moves qualification out of the browser page and into a testable Node reporter while retaining all browser-side raw samples
• writes JSON and Markdown reports with exact Git state, browser/machine identity, canonical runtime hashes, and the resolved PGlite package tree
• keeps quick mode as correctness/parity smoke evidence and makes it explicitly ineligible for performance claims

The old every-metric 30% rule was brittle for sub-millisecond operations and overweighted insert diagnostics. This gate is not a weaker semantic comparison: workload assertions, cross-origin isolation, PostgreSQL durability settings, and indexed-insert WAL parity must pass before either topology can qualify. Cold first open, duplicated workload-open time, close, and insert decomposition remain visible but descriptive; close is excluded because the public APIs make different worker-reclamation guarantees.

Full browser result

The qualifying run is bound to clean commit 3972f98c9666e83ee2b4a4f42d8389d9702b17ca and plan digest 89c2d4fe5c32c1d8593243578742404fcf7e8b4661cc40398ce953233b186493.

• Chromium 149, cross-origin isolated, 16 logical CPUs
• exact PGlite 0.5.4 source commit 25d0a55e1f1e4c59f26d9e125150dda88a33fd00
• exact installed PGlite tree b3925de04c386f51859c1bf18c143b225e3850616718140dd32e8eb48e9a2c84
• 5 startup rounds, 7 complete workload rounds, and 5 insert-diagnostic rounds with rotating engine order
• direct ratio: 0.5035 — 49.65% aggregate win
• worker ratio: 0.2029 — 79.71% aggregate win
• every primary metric's median paired ratio is below 1.0 in both topologies
• workload assertions, PostgreSQL durability parity, and WAL parity: PASS
• indexed-insert WAL delta: 24 bytes / 0.00123%, within the strict 0.1% tolerance

Representative metric	Oliphaunt direct	PGlite direct	Oliphaunt worker	PGlite worker
Warm open	172.87 ms	1,322.63 ms	1,084.89 ms	1,312.20 ms
Point-query median	0.055 ms	0.148 ms	0.225 ms	6.060 ms
Decode 10,000 rows	12.63 ms	40.64 ms	50.39 ms	91.81 ms
100-statement transaction	4.29 ms	12.68 ms	26.99 ms	480.70 ms

Reproduce with:

moon run oliphaunt-wasix-ts:bench-browser

or write to an explicit directory:

pnpm --dir src/bindings/wasix-ts package:build
node src/bindings/wasix-ts/tools/smoke-browser.mjs \
  --benchmark \
  --output target/perf/wasix-browser

Validation

• full clean-commit Chromium benchmark: PASS
• browser benchmark plan tests: 6/6
• Node benchmark plan tests: 6/6
• WASIX TypeScript tests: 19 files / 129 tests
• WASIX TypeScript typecheck and exact package shape
• SDK contracts and API surface
• repository policy checks
• pinned Biome and Cargo formatting checks

Base: #129 at 94854946681109552871f3d49aa2e740779882c5.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
oliphaunt-docs	Ready	Preview	Aug 13, 2026 9:50am