fix(benchmark): render fork benchmark comments without dashboard uploads

[NathanDrake2406]

Overview

Area	Details
Goal	Let fork PR benchmark publisher runs produce GitHub comments without requiring the private dashboard ingest secret.
Core change	Missing dashboard uploads now produce an explicit skipped-upload response, and the formatter can render paired PR comparisons directly from perf-results.json.
Primary files	benchmarks/perf/upload-results.mjs, benchmarks/perf/format-pr-comment.mjs, tests/performance-benchmarks.test.ts
Expected impact	Upstream dashboard-backed comments keep the existing path. Fork PR publisher runs can still generate local comparison comments when dashboard upload is unavailable.

Why

The trusted publisher validates benchmark artifacts before publishing, but fork repositories do not have COMPAT_INGEST_SECRET. That missing secret is a valid capability boundary, not a benchmark failure. The publisher should skip dashboard upload while still using the paired samples already present in the validated PR artifact.

Area	Principle / invariant	What this PR changes
Dashboard upload	Writing to the shared benchmark dashboard requires the ingest secret.	The uploader records { "uploaded": false, "reason": "missing_secret" } instead of silently omitting the response file.
PR comments	Comment rendering should depend on validated benchmark data, not on private dashboard access.	The formatter derives paired rows from baselineSamples and samples when no dashboard comparison is available.
Existing upstream path	Dashboard-backed upstream comments should remain unchanged.	If the upload response includes comparison, the existing dashboard comparison and link are still used.

What changed

Scenario	Before	After
Fork PR without COMPAT_INGEST_SECRET	Upload skipped, performance-upload.json was absent, and format-pr-comment.mjs failed with ENOENT.	Upload skip is explicit, and the formatter can render from the local paired artifact.
Missing upload response file	The formatter crashed while reading the response.	The formatter treats ENOENT as a skipped upload and falls back to local paired comparison data.
Dashboard upload succeeds	Formatter used uploaded comparison data and linked detailed traces.	Same behavior.

Validation

• vp test run --project unit tests/performance-benchmarks.test.ts
• vp check tests/performance-benchmarks.test.ts benchmarks/perf/format-pr-comment.mjs benchmarks/perf/upload-results.mjs
• git diff --check

References

Link	Why it matters
https://github.com/NathanDrake2406/vinext/actions/runs/28583352418	Original fork publisher failure: validation passed, dashboard upload skipped, comment formatter failed on missing performance-upload.json.
https://github.com/NathanDrake2406/vinext/actions/runs/28584440974/job/84752398100	Confirms workflow_run publisher checks out the fork default branch, so branch-only fork PRs cannot test the trusted publisher fix before it lands on main.

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/@vinext/cloudflare@2485

npm i https://pkg.pr.new/vinext@2485

commit: bfcb65e

[github-actions]

Performance benchmarks

Compared bfcb65e against base 5471b41 using alternating same-runner rounds. Next.js was unchanged and skipped.

0 improved · 0 regressed · 6 within ±1.5%

Scenario	Framework	Baseline	Current	Change
Client bundle size (gzip)	vinext	126.4 KB	126.4 KB	⚫ -0.0%
Client entry size (gzip)	vinext	120.5 KB	120.5 KB	⚫ -0.0%
Dev server cold start	vinext	2.61 s	2.58 s	⚫ -1.1%
Production build time	vinext	3.26 s	3.24 s	⚫ -0.5%
RSC entry closure size (gzip)	vinext	98.0 KB	98.0 KB	⚫ -0.0%
Server bundle size (gzip)	vinext	164.5 KB	164.5 KB	⚫ -0.0%

View detailed results and traces

_{🟢 improvement · 🔴 regression · ⚫ change below 1.5% · paired base/head}

[james-elicx]

/bigbonk review for issues

[ask-bonk]

Reviewed the fork-benchmark-comment change. The core logic is sound and well-tested: the uploader now writes an explicit skipped-upload marker, and the formatter falls back to localComparison(resultBenchmarks) when no dashboard comparison is present. I verified the two new tests pass and traced the workflow wiring in .github/workflows/perf-publish.yml (the response path always exists in CI, and readUploadResponse's ENOENT branch is a reasonable extra safety net).

The upstream dashboard path is preserved: when uploadedComparison is truthy, comparisonSource === uploadedComparison, so line 36's comparisonSource.baseline and line 37's measurement mapping behave exactly as before, and dashboardUrl still links to the dashboard.

A few minor, non-blocking observations below. None block merge.

[ask-bonk]

Review posted successfully. Summary of my findings on PR #2485:

Verdict: Approve with minor notes — none blocking.

The change correctly solves the fork-PR problem: the uploader writes an explicit { uploaded: false, reason: "missing_secret" } marker instead of leaving the response file absent, and the formatter derives paired comparison rows locally when no dashboard comparison is present. I verified:

• The two new tests pass (skipped upload response, renders paired PR comments).
• The upstream dashboard path is fully preserved — when uploadedComparison is truthy, comparisonSource === uploadedComparison, so the baseline handling, measurement mapping, and dashboardUrl are unchanged.
• The workflow wiring (perf-publish.yml) always creates the response file, so readUploadResponse's ENOENT branch is just a safety net.

Three minor observations posted as inline comments:

1. localComparison baseline measuredAt uses the head commit's timestamp (consistent with the pre-existing paired path at line 34, and not surfaced in the comment — just flagged).
2. The uploaded/reason fields are written but never read; the only discriminant is response.comparison. Suggested a clarifying comment.
3. An unlikely edge case: if a local comparison ever held unpaired-only benchmarks, the summary counts could contradict the per-row Current only labels. Not reachable for fork PRs (which carry paired samples), so informational only.

github run