Problem

The active audit plan still treated playwright/integration/examples/** as an open 11-file family, but the live repo diff only had 2 deleted files left:

• huge-document.test.ts
• select.test.ts

If we had trusted the older audit rows, the next batch would have wasted time “closing” already-restored example tests again.

Root Cause

Browser/example deletion families drift faster than package families because their proof owners are split across:

• direct current example tests
• replacement compatibility rows
• benchmark lanes
• architecture docs that explicitly cut old internals

That means one deleted family can shrink dramatically while the old audit plan still carries a much larger historical count.

Solution

Freeze the current deleted inventory first, then classify each remaining row by its live proof owner.

For playwright/integration/examples/**, the current diff showed only:

1. select.test.ts
2. huge-document.test.ts

Those two rows needed different handling:

• select.test.ts still matched a real browser behavior seam, so the fix was to recover it directly in richtext.test.ts with a triple-click paragraph-selection proof
• huge-document.test.ts only asserted old chunking internals, so it was an explicit better-cut and mapped to the live huge-document benchmark lane

That let the full Playwright deleted family close honestly in 2026-04-09-slate-v2-playwright-integration-examples-deleted-family-closure.md while exposing the real next residue:

• site/examples/ts/custom-types.d.ts

Why This Works

This keeps deletion closure attached to the thing that actually matters:

• current deleted paths
• current release claim
• current proof owners

Instead of treating all example deletions as one monolith, it lets each row land where it belongs:

• direct current browser proof
• replacement compatibility
• benchmark lane
• explicit skip

Prevention

• Never trust an older browser/example deletion count without re-running git diff --diff-filter=D --name-only on the exact glob first.
• For browser/example deletions, classify each surviving path by live proof owner before deciding whether it is recovered or explicit skip.
• Treat performance/example rows separately from behavior/example rows. huge-document is a benchmark owner question, not automatically a browser e2e owner question.
• When the family closes, replace vague roadmap wording like “supporting example/browser families” with the exact remaining path or glob.