Verification: one verbosity switch over the whole trace, defaulting to quiet

The verification node's record, against testing-plan.md. Every command below was run in this session from ui/ after bun install --frozen-lockfile.

Results

Row	Command	Outcome
T1	bun run test src/api/model.test.ts	pass — 101 tests (37 of them new: the isReadable row table and the isBookkeeping classification table)
T2	bun run test src/components/Transcript.test.tsx	pass — 18 tests (10 new, covering both views quiet and verbose, the hidden-count line and its singular)
T3	—	n/a as planned; the diff touches no route, schema or payload
T4	bun run test src/App.test.tsx	pass — the panel loads with the switch off and no tool group, and the switch restores them
T5	browser pass, below	pass — four screenshots committed beside this file
T6	—	n/a as planned
T7	—	n/a as planned
T8	bun run test src/api/model.test.ts src/components/Transcript.test.tsx	pass — an error in a hidden family is never classified as plumbing, and renders in the trail
T9	bun run test src/App.test.tsx	pass — role="switch", aria-checked tracking, aria-label no longer Tool calls, Enter and space both toggling
T10	bun run test	pass — 266 tests, 16 files. Nothing persisted changed; see the Python note below
T11	—	not run — no live service with a long-polled work item exists in this environment; the demo fixture's trail (T5) is the closest available and is committed
T12	markdownlint-cli2 over the changed Markdown	pass — 0 errors over 5 files

Also run, as the CI ui job runs them:

Command	Outcome
bun run lint (oxlint, type-aware)	pass, no findings
bun run build (runs tsc --noEmit first)	pass — 59 modules, 311.86 kB

The browser pass (T5)

The built app in demo mode behind bun run preview --port 4173, driven by Playwright against the pre-installed Chromium, one capture per state opened fresh. Counts read off the live DOM:

State	Item	aria-checked at load	Rows
trace-quiet-{light,dark}.png	loop-lab#214 (transcript served)	false	0 tool groups
trace-verbose-{light,dark}.png	same, after one click	true	2 tool groups
trail-quiet-{light,dark}.png	loop-lab#187 (no transcript)	false	0 event rows, and the hidden-count line in their place
trail-verbose-{light,dark}.png	same, after one click	true	1 event row (poll.spawn_failed)

The quiet trace screenshot is the acceptance criterion made visible: the human reply, the agent's two prose turns, the thinking disclosure, the malformed finding and the question banner, with no tool group between them. The quiet trail's single hidden row is a level=warning poll.spawn_failed — the documented trade-off in design.md § What it costs, seen working.

The switch returns to its default when another work item is opened, because the panel remounts per item. That is wanted: quiet is the default per item, not once per tab.

The Python side

make check could not run here: this container's uv is 0.8.17 and pyproject.toml requires >=0.12, <0.13, and uv self update is refused by a GitHub API rate limit. The change touches no Python — git diff --stat lists seven files, all under ui/ and docs/ — so no Python behaviour is in scope for this work item. The repository's CI runs make check on the pull request, and that run is the proof.