Review debt: book room was 1,245 draws GLB-on. Diagnosed two drivers —
multi-material stock boxes (one draw per face, 6x) and one mesh per stock
item — then batched:
- stock.js: stock boxes are single-material now; every stock mesh tagged
userData.isStock.
- batch.js (new): batchRoom() two-tier merge, called from interiors.js before
the GLB upgrade. Room-level merge of all isStock across every fitting
(grouped by material, world transforms baked) + per-fitting merge of fixture
frames (kept in-group so record bins stay raycast targets for ?dig=1 and the
GLB upgrade can hide the fixture). Uses vendored BufferGeometryUtils
mergeGeometries.
- glb.js: frame-hide is now by isStock tag (not frame-count), so batched
fixtures hide while the room-level stock rides on the GLB.
- interior_test.html: drawSweep() asserts <=350 draws across all type x
archetype (GLB off and on), folded into the soak; exposed as
window.PROCITY_C.drawSweep for F's harness (hook in LANE_C_NOTES).
Result (drawSweep, seed 1990): GLB-off worst 152, GLB-on worst 313
(opshop/hall) — book 1,245 -> 189. Determinism 0 fails / 810, leak-free, dig
still riffles batched bins, worst warm build 38ms (<50), visually identical.
qa.sh --strict GREEN (5 gates).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
35981ade5f