110 fittings placed across 12 shop types via new kit41.js (96-row catalogue, 67 ids in
pools) and ONE generic kitProp builder; recipes name pools, never asset ids. glb.js now
honours seeded per-fitting variants, multi-GLB booths, and a MEASURED autoYaw instead of a
110-row facing table.
THE INSTRUMENT WAS WRONG. drawSweep cleared every interior EXCEPT current, so every
historical interior number carried a phantom second room — 'toy', untouched this round,
'gained' 4 draws. R39/R40's celebrated 'worst 188, margin 162' was inflated. Corrected, and
with 110 new fittings placed: worst room in the game is 122 @ dept/auto, MARGIN 228.
Per type: record 61->78, pub 45->59, band_room 39->50, stall 71->82, book 51->61,
pawn 60->70, video 54->63, rsl 52->60, dept 116->122, opshop 102->105, milkbar 68->69,
toy 86->86 (+0, the control).
The pub finally has a bar (plywood_bar), banquettes, pokies, a stage flank. Milk bars get a
bain-marie and pie warmer. The pawn shop gets an 808 and an SP-1200 on the bench. Record
shops get a rigged DJ booth. Market stalls get 2-4 diggable bins each, closing R39's measured
44%-no-crate hole to 0/40.
WARDROBE, route R5: 257.5 KB total (246 KB atlas + 17 KB index), ONE fetch, one material,
one draw per fitting — 52 layers picked BY NAME, alpha-cropped, shelf-packed, 3.7% of source
bytes, lazy behind ?stock=real. LICENCE CALL MADE EXPLICITLY: AUDIT.md says no real brands
ship and the source library is prompted FROM real labels — the adidas jacket renders a
legible trefoil. Screened on a contact sheet, then relabelled in parody voice (Paddy's
Bootleg, Kaymart, Coogee Knits, Mombo, Stoosh).
FIVE DEFECTS FOUND AND FIXED, one long-standing: buyMesh has been INVISIBLE on every GLB
fitting since R9 (hiding every buyable item under ?stock=real) · DJ booths were
nondeterministic (parts attached in network order) · booth gear floated at 0.92 m · batch.js
dropped userData sub-tags on merge, so any post-batch flag read was already a no-op · the
path guarantee could empty a room (pub/wide/4242 returned a bar, a stage and nothing else) —
new restore pass, 129 pulled -> 101 restored over 480 rooms, path re-proved on every restore.
Gates: 288-room sweep x 4 arms all pass; soak 60 rooms avg 3.83 ms worst 10.3 ms, leaks 0;
960-room corpus 0 path-fails 0 carves; determinism byte-equal across fresh contexts
(sha256 c26b4848); ?noassets=1 60 rooms with 0 GLB / 0 manifest / 0 wardrobe / 0 stock
fetches. DJ control nodes recorded on fitting.controlNodes for a future driver (glTF drops
the limit constraints — a driver must clamp travel itself).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
422f17263e