diff --git a/.claude/tasks/plans/2026-11-17-v1-2-6-u2.md b/.claude/tasks/plans/2026-11-17-v1-2-6-u2.md index c3d3afb3..ee6876a2 100644 --- a/.claude/tasks/plans/2026-11-17-v1-2-6-u2.md +++ b/.claude/tasks/plans/2026-11-17-v1-2-6-u2.md @@ -1279,6 +1279,13 @@ Behaviour: of pointermove to next frame, loupe included, and the figures go into the plan. Over 16 ms at p95 with technique A, B is measured. If only B is acceptable, STOP for the conditional owner decision. The owner's tablet trial is the real check. + - Measured 2026-11-18 (`MEASURE_DRAG=1 MEASURE_BUILD=preview`, production build under `vite preview`, Chromium + touch 1 180 × 800, CPU ×4, 2 s drag of a « Vertèbres thoraciques » vertex on a seeded draft, three runs each). + Frame work (first `pointermove` to the end of the next frame), p95: technique A 4.2 / 4.5 / 4.7 ms, B 4.0 / + 4.4 / 4.3 ms, no loupe 4.0 / 4.2 / 3.6 ms; mouse control 3.1 to 3.9 ms. Technique A kept. From the event + timestamp the touch median is about 26 to 30 ms for all three (browser input queueing, the same without + loupe). The first run on the Vite dev server (React development build, StrictMode) gave about 30 ms for A and + 25 ms with no loupe; it does not stand for the product. - [ ] **Step 2:** FAIL. **Step 3:** implement. **Step 4:** gates. - [ ] **Step 5:** commit `feat(editor): two-finger pan, touch hint, loupe, touch-sized handles`.