docs(modeller): sync §OPEN LIST rows with their §SWEEP verdicts + rank what is actually left

[red1oon]

Housekeeping plus a ranked answer to "what's left on the Modeller".

The stale-claim trap I left

§SWEEP said "read my verdict column, not the 2026-07-30 status column" — but the table rows below it still read verified-open for rows this session closed. A row that still says open after being closed is exactly the trap this file exists to kill. The verdicts now live in the rows themselves:

row	now
19	✅ closed — move-gizmo recaptured, live bytes verified
23	✅ named D3 drift fixed (100% on all four discs); re-pointed at the live ELEC 2.11× over-count
24	✅ closed — e2e_harness.js:349 process.exit(), fixed in #1738
28	✅ does not reproduce — 293 seeded, 230 rotated, 0 dropped
32	✅ closed — #1704; witness re-run 8/0
35	✅ done today — site consolidated onto master and published

Ranked, so the next session doesn't re-derive priority

Row 36 first. IFC export handles only GEOM_EXTRUDE_POLY/GEOM_CUT/GEOM_ARRAY. Every ARC-seeded element is a GEOM_INSERT, so exporting an opened resident emits none of the building. Found by accident chasing row 34. Nobody has reported it — which suggests nobody has tried the feature on a resident.

Row 12 second. rel_fills_host ships for 3 of 8 residents, and dagevu_engine.js:62 builds every HostFillEdge from a real row — so #1706's anchor-by-default engine, the headline of the whole DAGeVu arc, is inert on 5 of 8 residents.

Two threads #1744 opened and did not close:

• The residual 11 — G4 honestly RED at 11 of 12,983, deliberately not relaxed. Suspects: the rx/ry guard (its premise is the same false claim corrected in #1738, and the path it distrusts is now witnessed — probably liftable) and the 1,301 of 3,225 elements with no resolvable blob.
• Datums 802 → 657 — plausibly a correction, but unverified. It bears on row 7's 0.104 m emergent-grid residual, since that grid is built from these datums. Re-measure row 7 before spending the heavy session it asks for — its baseline may have moved.

Sibling on record, unmeasured: compile_rooms.py reads the same coarse substrate. Exposure is not uniform — door_dims reads extents only and is safe (934/934); the position readers are exposed. Measure before touching.

⚠ Row 14's example doesn't reproduce — Dining_Chair is 0.443/0.427/1.227 m, upright. Item stands; re-source it or the witness is green on arrival.

🤖 Generated with Claude Code