Performance evidence

Measure a declared experiment

PliegoRS retains raw build and browser observations together with the exact revision, operating system, CPU, memory, storage declaration, Rust, Node, browser, sample count, cache policy, and known uncontrolled variables. p50 and p95 use nearest-rank without dropping outliers.

Build

Clean cold, no-change warm, content-only, CSS-only, and Rust-view observations

Browser

Signal updates, final DOM state, WASM linear memory, and mount/dispose residue

Reproduce before comparing

node scripts/measure-p8-builds.mjs
sh scripts/build-browser-benchmark.sh
node scripts/measure-browser-benchmark.mjs
node scripts/merge-p8-benchmark-report.mjs
npm run check:benchmarks

The merger requires both sections to name the same clean commit. A dirty smoke run may test the harness, but it is rejected as publishable evidence.

Pair speed with failure evidence

Performance does not excuse unsafe parsing or unbounded work. Fuzz targets and adversarial suites exercise manifests, receipts, graphs, paths, release bundles, telemetry state, content limits, and state restoration independently from benchmark timing.

Keep observations inside their boundary

The published P8 local baseline is useful for regression detection because it preserves its raw samples and limitations. Hosted candidate evidence remains a separate release gate.