docs: RELEASE-NOTES honesty pass — real 0.7.2 delta, historical counts restored

[matt82198]

The finding

RELEASE-NOTES.md's 0.7.2 section was fabricated — a near-verbatim copy of the 0.7.1 section's PR list. Every PR it cited (#638, #636, #643, #645, #647) shipped in 0.7.1, not 0.7.2. The one change that actually shipped in 0.7.2 appeared nowhere in it.

Both defects were introduced by the same commit, 2f52f39f (2026-08-02, "docs: RELEASE-NOTES/README currency to 0.7.2 with gate-derived test counts (266 tests)").

Ground truth: what 0.7.2 actually was

$ git log v0.7.1..v0.7.2 --oneline
e061f2bd Merge pull request #669 from matt82198/release/0.7.2
a00deb7f release: 0.7.2
a87c3966 Merge pull request #668 from matt82198/fix/api-state-empty-section-cache
fa29fb88 fix: /api/state served a collector's empty default instead of real data

Four commits. One substantive change (#668) + release metadata.

$ git show fa29fb88 --stat
 tests/test_agent_detail_roundtrip.py |  4 ++++
 ui/CLAUDE.md                         |  1 +
 ui/handler.py                        | 15 +++++++++++----

$ git show a00deb7f --stat
 CHANGELOG.md | 7 +++++++
 STATE.md     | 2 +-
 package.json | 2 +-

The section is rewritten to describe exactly that: the /api/state cache-staleness bug (mtime-gated CollectorSource seeded with {}; serve_api_state only computed inline when the cached payload was None, and {} is not None, so the endpoint served an empty data section as live state), the config-reload leak fixed alongside it, and the release-metadata commit. Fix detail is quoted from fa29fb88's own commit body.

Ground truth: the 0.7.0 test count

The 0.7.0 section's historical count was retroactively rewritten 254 → 266:

$ git show 2f52f39f -- RELEASE-NOTES.md
-3. **Test suite accountability**: 254 tests across 3 harnesses (25 Node + 13 Shell + 216 Python) ...
+3. **Test suite accountability**: 266 tests across 3 harnesses (26 Node + 13 Shell + 227 Python) ...

254 is the at-tag truth. The section was authored at 366a7fc4 as "254 tests across 3 harnesses (25 Node + 13 Shell + 216 Python)", and PR #644 ("docs: fix version drift to 0.7.0 and correct test count to 254") had verified that number for the release. Rewriting a shipped release's numbers to match a later, larger suite falsifies the record — a release note is a historical document, not a live dashboard.

254 is restored, with the currency concern handled by a footnote instead:

[count as of the v0.7.0 tag; suite has since grown — see current README]

(Nuance, stated plainly: the 0.7.0 section was authored during the 0.7.1 cycle — RELEASE-NOTES.md at the v0.7.0 tag only reaches back to 0.5.0. 254 is nonetheless the count as verified for that release by #644, and is the value the section shipped with.)

0.7.1 cross-check

Cross-checked against git log v0.7.0..v0.7.1 --oneline. Not fabricated — every claim traces to a commit in range. Two citations were wrong and are corrected, verified via gh:

Claim	Was	Verified truth
merge-train Unicode	#654	#654 is CLOSED. Change shipped as c6e465bf inside integration batch #667
portability-gate verdict	#667	502ccfaf; #667 is the integration batch PR (integrate: 0.7.1 batch — 12 PRs in one CI run), not the fix

Complexity-reduction claims were verified against PR titles and are now cited: #637 (_run_wave_inner F/141→C/13), #664 (wave_scheduler F82→C15), #658 (SSE collector), #662 (tracker reconciliation E39→C19), #663 (do_GET D30→A2).

All 13 PR numbers appearing in the touched sections were verified with gh pr view <n> --json title,state.

Scope

RELEASE-NOTES.md only. No gate in tools/, hooks/, or .github/workflows/ references this file. secret_scan.py --staged → CLEAN, exit 0.

🤖 Generated with Claude Code

[matt82198]

Evicted from the merge queue: batch #736 (integrate/q-1785730980) red with every member individually green

[matt82198]

Evicted from the merge queue: batch #766 (integrate/q-1785803317) red with every member individually green