Repository navigation
Conversation
chh-ay
added this pull request to stack #69
October 6, 2026 18:23
The performance showcase's decision record reads results/interaction-results.json, but that artifact was a one-off before/after pair of the 0.3.0 comparison with no producer: bench:release never refreshed it, and its improvement-ratio gates cannot be re-measured at a later release. bench:interaction now measures the current values of the four published metrics on the current build - packed inverse-index lookup latency and backing bytes, retained bytes after 100 distant edits, and the cold-route long task owned by Sheetwrite - in fresh worker processes and a fresh browser context, and records raw samples, their summary, runner provenance, and the method. Schema version 2 gates those metrics on absolute release ceilings, so no baseline build is re-measured and no old sample is carried forward. Cold-route ownership is reconstructed from a V8 CPU profile inside each long task and stated in the artifact. bench:release and its smoke rehearsal now capture the interaction and paged artifacts, and the bench and guide documentation describe that set.
Replace the 0.3.0 before/after artifact with a capture of the current build at the harness commit: five samples of packed inverse-index latency and backing bytes, retained bytes after 100 distant edits, the Sheetwrite-owned cold-route long task, and the cold route's usable time, each with its summary and the runner that produced it. Every release ceiling passed, and the version 1 pair now lives in git history only.
The performance evidence page is generated, so the capture set it names has to come from scripts/docs.ts: the page now states that bench:release refreshes render-scale, the showcase's paged storage, the interaction evidence, data, core paths, formula, the engine captures, and XLSX, and what the interaction capture measures.
The capture wrote the raw Chromium build number, so the artifact's runner provenance did not say which browser it was. Record `Chromium <version>`, as the other browser captures do; the next capture picks the string up.
The showcase now reads the interaction-current-v2 artifact: the four cards show the current value of their metric and its release ceiling, and they show the change only against a recorded predecessor capture, so a first capture states the current value instead of an invented before. The scenario module annotates the artifact shape it consumes, which keeps the predecessor optional.
chh-ay
force-pushed
the
bench/release-evidence
branch
from
October 7, 2026 08:39
3464279 to
dbad27e
Compare
86 of 96 tasks
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
paged-results.jsonwas also not part ofbench:release.bench:interactionmeasures the current build: lookup latency (median and p95), inverse view-index bytes, bytes after 100 distant edits, and owned cold-route long tasks. Long-task ownership comes from CDP traces and the V8 CPU profile.interaction-current-v2with absolute release ceilings: lookup p95 1,000 ns, inverse index 4,000,000 bytes, 100 distant edits 1 MiB, owned cold long task 50 ms. A comparison is shown only when a previous capture of the same schema exists. No "before" value is invented.bench:releasenow also captures interaction and paged evidence, with smoke variants. The regression baseline re-record stays last. The paged artifact now records its commit stamp.Verification
bench:release:smokepasses under the timing lock (render-scale, interaction, paged, data, core-paths, formula, xlsx).docs:check, and the performance browser spec (5 pass) pass. Layer check on the stack passes.