Skip to content

Session handover — 2026-07-14 pt4

A follow-on to pt3. The fcr.gallery body field (pt3's agreed next slice) shipped — WCP's contentDestroyed veto is cleared, and WCP now HOLDs on videoDebt alone. One commit: fa16347. WCP worker 1a9d6a45 deployed.

Where the two sites now stand

site verdict why
WCP HOLD videoDebt only (live references 27 videos, we serve 1). Composite 87.2 — S 82.2 / T 100 / G_det 84.4, C 100 (0/145 text-bearing sections destroyed).
garvanbay HOLD criticalSectionDropped (/contact has no hero) — unchanged, untouched all session (regen A/B: 6 diffs, all body: "", render-inert; worker not redeployed).

The gallery body slice (fa16347)

See the closed known-issues entry ("~~fcr.gallery HAS NO TEXT FIELD~~ — CLOSED 2026-07-14 pt4") for the full account. The headline: the filed fix chain was half already built. The prescription said matcher → schema → transformer → adapter → block; measured against the code, Gallery.js had extracted the intro as subtext all along, the static lane rendered it, and gallery-wcp's SectionProse slot needed nothing. The destruction was three CMS-lane gaps (fromGallery never read subtext; no schema field; adapter passed nothing), closed as the fcr.cta-strip.body shape verbatim.

  • Landed by surgical sync-style-to-d1 stamp (1 block patched, 12 preserved — never a reseed) + committed-seed patch, after a --dry pass.
  • A/B regen proof (change applied vs stashed): WCP 13 diffs, every one fcr.gallery.body (home = the 202-char intro, the rest "" — split-continuation galleries get no subtext by construction, faithful). garvanbay 6 diffs, all body: "".
  • Re-measured on the deployed worker (fresh ours re-capture, 25 pages): contentDestroyed GONE, C = 100, vetoes = [videoDebt]. S/T/G_det byte-identical to pt3 (82.2/100/84.4, composite 87.2) — correct, because the veto was the instrument, not the score.
  • Zero ui-missing placeholders across spot-checked pages; both intro sentences render live.

No new known-patterns entry: the slice exercised catalogued machinery ("The instrument that ALREADY EXISTS…", "One emitter, two item shapes" detection rule, the A/B-stash proof, the surgical stamp). The "check which lane actually drops the data before scoping the whole chain" lesson is those entries applied.

  • Home fcr.faq heading is cta copy — heading-EXTRACTION matcher bug; still blocks the 20th FAQ tone fix and still causes home's faq dropped S row.
  • Probe gradient-parsing — moot on both references; trigger = a site whose LIVE bands are gradient-painted.
  • apps/cms/src/components/blocks-v3/ orphaned pre-workspace copy (chore).
  • garvanbay's worker still carries the pre-1d bundle (provably no-op there).

Next

  1. Video segmentation/render workstream — now WCP's only HOLD reason. videoDebt carries the per-page id punch-list (gallery 13, home 4, …); segmentation (which section carries which video) is the work — pulling files without section homes is motion without progress.
  2. garvanbay /contact hero — its HOLD, untouched for several sessions.
  3. Still open from earlier: aligner role-aware anchoring (the hero: dropped 5 S residual), Slice 6 calibration, operator editability (upstream emdash save race).

The through-line

pt3's arc was measure the instrument before believing it. This session is the quiet corollary: measure the filing before scoping it. The known-issues entry prescribed a five-layer fix; the code said three small ones. Half a session's scope evaporated under one exploration pass, and the A/B-stash + surgical-stamp discipline landed the rest without touching either reference site's score — only its verdict.