[Perf] LHCI regression detected — run 24550351252

[github-actions]

The nightly Lighthouse CI run failed against the production budget defined in budget.json.

Run: see the "Performance" workflow in the Actions tab for the failing URL(s) and the specific assertion that breached.

LHCI report: linked in the workflow run summary (uploaded to LHCI temporary public storage — link expires after ~7 days, so click through soon).

Investigate

1. Check CLAUDE.md → Known Issues & Learnings → Performance & Web Vitals.
   The 13 documented incidents cover most failure modes. Walk the
   "Performance Troubleshooting Checklist" in the same section before forming
   new hypotheses.

2. Cross-check field data:

   • Vercel dashboard → Speed Insights tab (real-user p75 mobile/desktop CWV)
   • Microsoft Clarity → URL performance export

3. Run locally:

   • npm run perf-audit (production)
   • npm run perf-audit -- --compare (run-over-run delta)

Common causes

See the Performance Troubleshooting Checklist in CLAUDE.md — items
1 through 5 walk through the usual suspects in priority order.

When you fix it

Add the new lesson to the Performance & Web Vitals table in CLAUDE.md
before closing this issue. That's the contract — every new failure mode
gets documented so the next investigation starts smarter.

_{Auto-opened by .github/workflows/perf.yml. Close once resolved.}

[imsaif]

Closing as stale — the condition was resolved in April 2026, though the specifics are no longer recoverable.

What this issue actually contains: nothing diagnostic. The body is the static .github/perf-issue-template.md, which records no failing URL, no breached assertion and no measurement — only a pointer to the Actions run. That pointer is now dead:

failed to get run log: HTTP 410

GitHub expires workflow logs at ~90 days and this run is from April; the LHCI report link expired after ~7 days, as the template itself warns. Only metadata survives.

Why it's safe to close anyway: the Performance workflow went green on 2026-04-23 and 04-24, days after this issue opened, and stayed green through 2026-08-10 — 139 successful runs. Whatever breached here was fixed shortly afterwards. That fits the history: April 2026 was a heavy perf-fixing stretch and .claude/rules/performance.md documents ~11 incidents from that month (adjustFontFallback, carousel priority mis-promotion, <video> without preload="metadata", the 15MB animated webp, ssr: false above the fold, and others). The siblings from Apr 20–22 (#30, #31, #32) were closed at the time; this one was missed.

Deliberately not attributing this to #74 — that fixed an unrelated August budget breach and had nothing to do with this.

Root problem, now fixed: the template embedded no measurement, so these issues become unfalsifiable once their logs expire. The workflow now writes the failing URLs and assertions into the issue body at creation time.