V51: a store that fits on a screen

[dapiced]

Summary

Refactor with no intended behaviour change. The 906 existing unit tests and 97 Chromium e2e tests pass unchanged before and after.

Before	After
lab-store.ts, 1 050 lines	530 lines: actions only, public names re-exported, so no importer changes
onmessage: a 250-line if / else if chain	store/responses.ts: a dispatch table typed as a mapped type over WorkerResponse['kind']
State shape, V47 fence and threshold artifact inline in the store	store/state.ts, store/fence.ts, store/artifacts.ts
Test files v35 … v40	honest-evaluation, imbalance-ensemble, knn-and-helpers, french-numbers, column-recipe, quality-score
numbersOf / missingRatio / shares duplicated in drift.ts and reference.ts	one copy in quality/distribution.ts

Dispatch table

• A missing or extra handler is now a compile error. The old chain's final else used to treat any unlisted kind as error without saying so.
• The handler bodies were moved by script, not retyped.

Not done

• Feature slices: the actions share one worker and one send, so slicing them would move code without separating anything.
• Two quantile functions: left as two. On an empty list one returns 0 and the other NaN, so merging them would change behaviour.
• Mixed test files v35 and v36: renamed, not split, to keep the change mechanical.

Coverage gate (V49) fired, and the fix was a test, not a lower floor

• Moving the code turned 35 branches into 35 functions. The function denominator went from 2 142 to 2 174, and the 54 % floor failed at 53.72 %.
• I added store/responses.test.ts: 17 cases that pin which fields each state-only response writes. That is also the check a moved body needs.
• Function coverage is back at 54.46 %.

Test plan

• lint, typecheck, format:check, build
• npm run test:coverage: 923 passed (906 unchanged + 17 new), all floors met
• Playwright Chromium: 97/97

🤖 Generated with Claude Code

https://claude.ai/code/session_01ELAjHYDyxa5zbbdcke1bMA

---

Generated by Claude Code