docs: stop hardcoding test counts, and credit the PR that actually landed

[oratis]

Two defects found while reviewing the #23–#30 batch. Both are already on main, and both are the same class of rot that batch was written to remove.

1. Wrong PR credited

.gitattributes (from #29) and docs/memory.md §6 both say #27 fixed the usage-ledger.js NUL bytes. #27 was closed as a duplicate — #30 landed the fix.

I confirmed the two branches were byte-for-byte identical before either merged (both produced the same 10789-byte file, 0 NULs), so this was pure parallel duplication, not two different fixes. Following the reference today lands you on a closed PR with an empty diff — for a file whose entire problem was being un-reviewable.

2. The count treadmill

#26 corrected five hardcoded test counts (234 / 377 / 656 → 678). #28 merged minutes later and added a test. So the PR whose whole purpose was de-staling docs shipped a number that was already wrong:

moment	real count	docs said
#26 authored	656	656 ✅
#23 + #24 merge	672→678	656 ❌
my #26 review fix	678	678 ✅
#28 merge	679	678 ❌

Five copies of a figure that moves on every test-bearing merge cannot stay right. Corrections don't fix that — they just reset the clock.

What changed

• Counts are now a magnitude pinned to a commit (~680 tests / 71 files as of c7c7d5b) with an explicit instruction to measure rather than cite
• The two places that only ever needed "all green" (§7.3 deploy checklist, the serialized-run comment) no longer carry a number at all
• memory.md §5.2 records the anti-pattern, so the next doc pass doesn't reinstate it

The prose that never carried a number — "Vitest files under client/src/{components,pages,utils}/" — stayed accurate through this entire batch. That's the argument.

The surviving 234 / 656 mentions are deliberate: they're the history being explained, not live facts.

Test plan

• node --test --test-concurrency=1 server/__tests__/*.test.js → 679/679
• grep confirms no stale hardcoded count and no bare #27 reference survives
• Docs-only — no code path touched

Note on CI: a default npm test can show 2–5 red files with { code: 'SQLITE_BUSY' }. That's the shared-influencex.db concurrency flake documented in memory.md §5.2 — it reproduces on main with zero changes. Serialized is the triage step.

🤖 Generated with Claude Code