B-Spline progress

Seat A ↗ main · turn 363

advisor (reviewing)

task 31/35 · 89%

Item 14 NOT fixed this session (d458e78) -- capacity report, not a stall. Root cause (331deg wrong-branch arc sweep in Template 10's shared top-arc construction) was already precisely identified in item 13; this session tried 4 distinct live-Fusion fixes (exact sagitta seed, explicit Radius constraint, horn/arc seed consistency, constraint reordering), each disproven with measurements. Key finding: an isolated Fusion-API unit test (same seed, same Symmetry+Tangent constraints) reliably gives the CORRECT small arc -- rules out the math/formula/API, points to the full sketch's own pre-existing content (skeleton pins, BB projections) shifting which branch Fusion's global nonlinear solver lands in. Also found a reusable gotcha: TemplateLoader caches phase modules per Fusion session, so phase-file edits need an add-in stop/run to take effect, not just a fresh build call. Recommended next step (not implemented): solve the arc in an isolated temp sketch at build time with live numeric values, then feed the resolved radius into the main sketch -- stays parametric, sidesteps the branch-selection problem, but needs get_block() restructured to do live Fusion work, a real design change. Reverted all experiments; deployed add-in confirmed back at the clean committed state, no drift. 5 live-Fusion round trips this item alone. Recommend either a fresh session with more room for the isolated-pre-solve redesign, or your own call on priority/approach now that the depth is understood. Full detail in LIVE-RESULTS-ranchy.md and WORK-LOG-fb-app.md.

updated 2026-10-01T04:35:32Z

+54 more, swipe in the viewer

Seat B ↗ lane-b · turn 195

worker (working)

task 26/27 · 96%

epoch 8 - inset window app side reviewed OK. FIRST, a quick one: lane-b still carries the REJECTED Template 7 (e2ab242 + its follow-ups: template_7/ folder, T7 frame-defs, PARAM keys, tests). Remove it cleanly so lane-b can merge to main without shipping it: 'git revert' those T7 commits (or one commit removing template_7 + its keys/tests + regenerating frame-defs), keep the inset-window commits intact, full suite + A/B T1-6,8,9 byte-identical, push, and PASS BACK right away (I merge the inset window then). THEN T7 from scratch per the approved shape: miter diagrams at 7x9 + 6x9 for Fred first.

updated 2026-10-01T02:43:59Z