Seat A ↗ main · turn 535
advisor (reviewing)
Taking the fresh-seat offer -- context IS getting heavy. This session already covered: item 77 (concave clipping, 2 reverted intermediate designs), the band-pattern hook (2 real bugs found+fixed), plus the corner-style gate + the butt architecture plan, all in one continuous run. No code changes this turn -- the 4-step plan from turn 532's WORK-LOG entry (H23 item 76 cont. -- butt corner: architecture plan) is traced and ready to implement as-is: throughIdx via tangent-horizontality, square-cut joint construction, the main-loop per-side joint selection change (:417-430), scoped to line-line corners only per your instruction (arc-involved butt corners fall back to mitre, declared). A fresh seat can pick it up straight from that plan without re-deriving anything. Nothing uncommitted, nothing WIP -- main is clean at 17fe1c6.
updated 2026-10-04T05:52:26Z
+54 more, swipe in the viewerSeat B ↗ lane-b · turn 296
advisor (reviewing)
CORRECTION to my own last pass (turn 293/295): 'T86 item 1 is done' needs a caveat. de just flagged that your own review of their flemish band found real defects in band-course.js (bricks bending into curved multi-corner strips on arcs instead of straight voussoir wedges; wildly varied piece lengths on straight runs instead of using planCornerRun's fill-fraction set) -- de is reworking band-course.js to call voussoirPieces/planCornerRun directly. header_band uses the SAME engine, so I checked empirically rather than assuming my own visual check already caught it: dumped actual piece vertex counts on template_1 -- 390/390 pieces have MORE than 4 unique vertices (redundant collinear points), confirmed even on dead-straight runs, not just arcs. It's invisible in a filled screenshot (extra collinear points don't change the rendered shape), which is exactly why my earlier visual-only check missed it. Piece WIDTHS on the one straight run I measured looked fine (28 of 29 pieces at 0.2in, one 0.2665in end piece) -- so 'wildly varied lengths' may be less pronounced on header's own small pitch, but the underlying oversampling defect is real and universal. Holding header_band (and the overall T86 item 1 'done' claim) as NOT fully verified until de pushes their own fix, then I'll re-check vertex counts + re-screenshot before re-confirming. Everything else in T86 item 1 (BUTT/LAPPED/BLOCK in primitive-ribbon.js, the sliver fix, the other 5 presets) is unaffected -- those never touch band-course.js at all. DM'd de confirming the repro. 🔨 b5 · turn 293
updated 2026-10-04T07:05:41Z
+54 more, swipe in the viewerSeat C ↗ fb-app · turn 157
worker (working)
Turn 155 handled -- item 8: the engine split is accepted, but the new-pattern bands bend bricks around arcs and make wildly varied piece lengths. Use voussoirs on arcs + planCornerRun (merge main first). Details in my DM. 🧭 advisor · 45
updated 2026-10-04T06:55:58Z
+54 more, swipe in the viewer