Speed up compatibility matrix lookups

[jonbaldie]

Summary

The /compatibility matrix mounted a hidden tooltip for every badge (~32k elements with all 3,662 rows expanded) and re-rendered every badge on each search keystroke.

 <NodeCompatibilityMatrix>
-  <StatusBadge>                  # rerenders on every keystroke
-    <span role="tooltip">        # always mounted, CSS-hidden
+  <StatusBadge> = memo(...)      # skipped when props are unchanged
+    {hovered || focused ? <span role="tooltip"> : null}
-  <StabilityBadge>
-    <span role="tooltip">
+  <StabilityBadge> = memo(...)
+    {hovered || focused ? <span role="tooltip"> : null}

Every row, label, filter, and tooltip text is unchanged. Tooltips still appear on hover and keyboard focus.

tests/browser/
├── compatibility.test.ts     # 4 Playwright contracts, run by docs `pnpm check`
├── compatibility.bench.ts    # timing benchmark (opt-in, never a CI assertion)
└── compatibility-oracle.ts   # independent expected rows from the published census
internal/performance/compatibility-lookup.md   # method, samples, experiments, reproduce steps

CI gains a required test (docs compatibility browser contracts) job that runs the docs gate with Chromium.

Evidence

Lab-only: production Next.js build, Apple M3, Chrome for Testing 154, 9 attempts per phase. Measured from first input until two animation frames after the final matching readFile result.

CPU	p50 before → after	Max before → after
Normal	1,087.7 → 514.2 ms (−52.7%)	1,491.0 → 556.9 ms
4× throttle	3,799.8 → 1,972.4 ms (−48.1%)	4,644.6 → 2,101.2 ms

• Before: an inactive expanded matrix has hidden tooltip nodes, and r → re calls the status renderer 6,128 times. The new contracts failed on 2069684.
  After: 0 tooltip nodes, at most 17 × rows + 500 elements (58,649 vs 90,922 before), and ≤ 16 retained-badge renders per kind. All 4 contracts pass.

all APIs remain reachable without eagerly mounting hidden tooltips        ✓
search and independent tier filters preserve every match and its parents  ✓
retained API badges do not rerender on every search keystroke             ✓
status and stability details remain available to mouse and keyboard users ✓

Settled tooltip screenshots: desktop is pixel-identical, and mobile differs in 92 of 912,600 color channels.

	Before	After
Desktop
Mobile 390×844

A CSS content-visibility experiment was rejected because it collapsed table height from ~210k to ~68k px, changing scroll geometry.

Validation. On this branch (rebased onto 0845c7f), both frozen installs and NEXT_DIST_DIR=.next-check pnpm check pass. pnpm test:sandbox was unavailable (expired Vercel credentials). The local fallback ran on 2069684 and is not green:

• plain: 7,267 passed, 1 failed. zig COFF dry-run inputs retain complete cross-target cache hits failed with ENOENT scanning cache/bin after 34.7 s under load. It passes in 5.9 s when run alone.
• san: 8 failed. Every one is the sanitizer printing No external symbolizers found … Is PATH set? into byte-compared stderr. It reproduces alone via 2946-child-fork-ipc, which forks with env: { IPC_FIXTURE: "yes" } (no PATH). Upstream's macOS 15 san shards passed on the same base commit, so this appears specific to the Darwin 25.6 host.

This PR changes no compiler, runtime, or harness code.

Merge Danger

Door: two-way

Revert the commit to restore the old matrix and remove the CI job.

Blast Radius: docs-and-CI

• Tooltip content now mounts on mouseenter/focus. Settled visuals match, but the hover/focus fade-in has not been checked on real devices.
• Adds @playwright/test as a root dev dependency.
• Adds a required CI job (~Chromium install + docs build) to the test aggregate. A flaky browser would block merges.

🤖 Generated with Claude Code

[vercel]

@jonbaldie is attempting to deploy a commit to the Vercel Labs Team on Vercel.

A member of the Team first needs to authorize it.

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	npm/​@​playwright/​test@​1.63.0

View full report