Session handover — 2026-09-02 pt4

Continues pt3. Ran past midnight — the last commit lands 2026-09-03 11:32.

No slice was scoped. This part was driven entirely by the owner's questions, and the pattern is worth noting for whoever reads it: five of the seven answers changed a wiki claim or found a defect, and two of them corrected claims this session had written hours earlier. The questions were better than the plan.

Slice 6 is still waiting on human labels — six sessions now.

Where to look

known-issues — the new material is clustered around the CMS/editor section:

• "What every admin sidebar item is FOR" — the map of all eleven admin screens, measured by counting rows on both live databases. Two carry our content; eight are empty; one is a trap. Read this before anyone shows the admin to the team.
• "SPIKE RESULT — garvanbay's footer now runs on emdash's native Menus + Widget areas" — including the header hybrid, why WCP's split header still works, and the dropdown-flattening mistake.
• "Analytics & marketing tags" — the fcr.analytics block, the crawl-side capture, and the indexable gate.
• "Adding Wix-style JS transitions" — feasibility assessment (asked for understanding, not build) plus a dead-tree trap in the CMS app.
• ""Email — install plugin"" — what that message means and why we probably need no plugin.

known-patterns — two new entries, both from mistakes made here: assessing a platform feature in isolation, and verifying a chrome change by diffing markup rather than counting elements.

What shipped

commit	what
aa6f7b1	the admin map — every sidebar item, measured
9b13a16	hiding sections is impossible (config surface enumerated); the nav recommendation
a864dcc	menu delete held on the owner's call; email conclusion corrected
d82bcfc	fcr.analytics — GTM/GA4/Ads/Meta/TikTok/verification, IDs only, consent denied
57a63a5	Wix-transitions feasibility
9927e91	correction: Widgets are not redundant — they are emdash's region model
8781d74	spike: garvanbay's footer on native Menus + Widget areas
298583f	spike part 2: header nav hybrid; decoy adopted, not deleted
51387e6	the crawl now captures the client's marketing tags

Deployed: garvanbay 0591bb13, WCP 69261d14 — both on the same code, verified byte-identical header and footer, 0 tag output, still noindex.

The honest parts

• Two claims written earlier in this same session were wrong and had to be corrected. "Widgets and Sections are genuinely redundant" (half false — Widgets is the region model), and "G-20251107 looks like a date" (it is part of an image filename, IMG-20251107-WA0117.jpg). Both corrections are in place with the evidence; neither was caught by a check, one was caught by the owner and one by writing a stricter regex.
• The header spike shipped a degraded header to the live site for a few minutes — the "More" dropdown flattened to a plain link. Caught by a byte-diff, fixed, and the lesson banked. A count-based check would have passed it.
• Live databases were written to repeatedly — media rows, widget areas, menus, the site title. All additive or reversible, all recorded, but this part of the session touched production data far more than a spike normally should.
• tags.json is gitignored like other build artifacts, so the two files created here are local only. A re-crawl regenerates them; the 45 corpus sites have none yet.
• The team still has not touched any of it. The footer/menu spike exists to answer "is this better for the 40 people who use it" and that question is still open. Everything recorded about the editing experience is one person's read of a UI.
• Nothing here advanced Slice 6.

Next

1. The labels. Unchanged, unblocked, and the only thing on the critical path: calibration/review.html → label the 15 SHIP candidates → calibration/labels.csv → fit-calibration → promote only if fitted:true and precision holds.
2. Put the native footer/menu in front of two or three of the website team and record what they say. That is the spike's actual deliverable and only they can produce it.
3. Editor audit on WCP — asked for three handovers ago, still not done. Its carousel, hero video and split-center chrome are unexercised by garvanbay.
4. Two parked decisions: the decoy menu row (held deliberately — decide it with the fcr-vs-native question, not separately), and wiring the cookie banner so analytics can record anything at all.
5. Re-crawl to capture tags for the corpus — no code needed, but nothing has run it.