Session close: preserve the repositioned copy, sync the ledger, record the one red

.okf/log.md

@@ -51,6 +51,28 @@ make it green: restructure same-day entries under one heading, and add
 `timestamp` to the 23 concepts missing it (anchored to each file's last commit
 time, which is verifiable - never invented).
 
+## 2026-08-22 - a re-record that fixed 21 of 22, and the one it could not
+
+#566 merged: 127 baselines re-recorded and the tolerance shield deleted. CI
+Linux red went from 22 keys to 1. The residual, `mobile/blog/index/_pagination`,
+is worth its own note because it defeated two wrong diagnoses in a row.
+
+**It is not flake.** Two runs of the same job returned difference_level
+0.03960763888888889 to sixteen decimals. This file's own rule says identical
+difference_levels across runs mean a stale committed baseline, not flaky
+rendering - and that rule held, against an intuition that said otherwise (the
+key had just been re-recorded 35 minutes earlier, which FEELS like flake).
+A run-level `conclusion: success` was also misread as a passing re-run; the
+job-level conclusion was `failure`. **Read the job, not the run.**
+
+**What it probably is, unverified:** the `workflow_dispatch` record path and
+the `pull_request` test path build the site differently, so the baseline was
+captured from a page the test run never renders. Blog-index pagination is
+driven by the post count, which is exactly the kind of thing a draft-handling
+difference would move. Whoever picks this up: diff the two runs' build steps
+before touching a baseline. Re-recording it again without that answer just
+buys the same failure at a different sha.
+
 ## 2026-08-21 - three register pilots shipped; the comparison is controlled; Editorial recommended
 
 All three register pilots merged on the /next/ rail (#561 rescue-room, #562

docs/projects/2608-site-design-system/20-29-strategy/20.06-clean-slate-v2-plan.md

@@ -157,7 +157,10 @@ same commit · browser-measured AA + seams on anything touching a rendered page.
 | Pilot C (Engineered Precision) | **MERGED** #563 (`45780ad6f`); DOM-verified copy parity across all three | — |
 | Lightning demo | **BOARD LIVE** — `20.08`: pilot vote B 2/C 1/A 1; artifact head-to-head Editorial 4–1; fable rec = Editorial + graft package | **PAUL PICKS** on the live /next/ URLs |
 | Visual-gate repair | #560 merged (`e1882483c`): tolerance 0.0001 (was 50x blind), shield documented, fossilization + viewport-only + perceptual-floor documented; machine exonerated (reds = #540-stale baselines) | follow-ups: re-record ~36 macOS + 22-key Linux dispatch; below-fold coverage |
-| Homepage build (Editorial) | Blueprint ready (CD `8eb98dd6` Homepage.dc.html, team section removed) | GATED on Paul: audit-exhibit sourcing (real/SAMPLE/rebuild) + "code audit" offer naming |
+| Homepage build (Editorial) | Blueprint ready (CD `8eb98dd6` Homepage.dc.html, team section removed) | GATED on Paul: audit-exhibit sourcing (real/SAMPLE/rebuild) + "code audit" offer naming + its "Three ways in" section contradicts one-offer positioning |
+| **POSITIONING CHANGED 2026-08-21** | Site sells ONE thing: an embedded team of senior self-managed full-stack developers. **No fractional-CTO / tech-lead title claims** (factual-accuracy rule). Pilot copy rewritten + canon-verified but NOT APPLIED - `20.09` | FIRST TASK NEXT SESSION: apply 20.09 to all three stubs, rename to `embedded-team`, re-record 6+6 baselines |
+| Visual gate: baselines | **#566 MERGED** (`6420e644e`): 127 baselines re-recorded (50 macOS classified by pixel-transition histogram, 77 Linux CI-recorded then screened), SECTION_CONFIGS shield deleted on a measurement. macOS suite GREEN - first time since #540. CI Linux red went 22 keys -> 1 | see the residual row below |
+| KNOWN RED: `linux/mobile/blog/index/_pagination` | Fails at difference_level **0.03960763888888889, IDENTICAL across two runs** - so per CLAUDE.md's own tell it is a stale/wrong committed baseline, NOT flake. It was already red before #566 and the CI re-record did not fix it; root cause NOT diagnosed (suspect: the `update-baselines` dispatch and the `pull_request` run build the page differently - e.g. draft handling changing the post count that drives pagination) | own investigation; do not whack-a-mole re-record |
 | PR-1: v2 rail | **MERGED** #555 (`703fee12c`) — rail live: `/next/` noindexed staging, qtest holes closed (root layouts + v2 registry-derived keys), ADR-0006, mermaid source-aware. Screenshot job red = the documented 16-key #540 handoff only, verified key-by-key. | — |
 | PR-2: styled pilot | **MERGED** #557 (`38f744805`) — full content ported (overview/faqs/tenure-derived stats), styled per D1, canon-exact Clutch proof line, browser-verified AA (button 5.13:1), −85.6% gzip CSS vs legacy page. Screenshot red = the same 16-key #540 handoff set, verified key-by-key. | compare links: `/next/services/fractional-cto/` vs `/services/fractional-cto/` |
 | PR-3: flip | BLOCKED on lightning-demo winner + Paul's live review | flip = `type: next` on the real page + delete stub; checklist in ADR-0006 §5 |