Testing plan: de-cluttering the control plane UI (Option C — Minimal)

Derived from requirements.md and design.md — the C columns of its component table and element inventory — before tasks.md. Authored at test-planning, completed at verification.
This file is executable content. It names commands an agent will run. Credentials appear by reference only — none are needed here.

Test matrix

The proof is three-fold: the component suites assert each inventory verdict and every preserved action and state; headless screenshots of the built app prove the rendering against the locked prototype in both themes; the CI commands and a bundle comparison prove the non-functional requirements. Everything runs on the demo fixture — no service, no credential.

#	Type	Applies?	Scope / what it proves	Where it runs
T1	Unit	yes	`railNote` no longer names the node; `sidebarGroup` still answers four ways while the sidebar maps to Needs you / Everything else sorted by last activity (`SIDEBAR_GROUPS_MINIMAL`); `compactRail` and `railLabel` unchanged; `filterViews` unchanged; `rowFlag` unchanged; `chatBlocked` unchanged	`cd ui && bun run test`
T2	Integration (scenario)	yes	the React suites on the demo transport (`App.test.tsx`, `Sidebar.test.tsx`, `GraphStrip.test.tsx`, `WorkItemDetail.test.tsx`, `Transcript.test.tsx`, a new `SessionFacts.test.tsx`): every inventory verdict for C — what each band renders by default and what a disclosure reveals; every action of issue-327's functionality map still reachable (R5.2); every state still rendered (R5.3); the one-session tab rule; the single statement of the current node	`cd ui && bun run test`
T3	Contract (OpenAPI)	n/a — no route, request or response shape changes; `ui/src/api/`, `ui/src/state/` and `ui/src/demo/` are byte-identical (R5.1), asserted by T12's diff check
T4	End-to-end	n/a — no live service in this environment; the demo transport is the same client interface over the same record shapes, and nothing in the data path changes
T5	UI / visual	yes	the built app, demo mode, headless Chromium at 1440×900 and 390×844, both themes: Work with #214 selected; PR #216 session (the `work item` kicker beside the verbs); Full graph on; the gate banner (#205); the question banner (#214); Session details open and closed; the search icon open; the no-transcript fallback and blocked composer (#187); Standing; Settings; the sidebar collapsed; 390 px with the sidebar closed and with Session details open — each matched to the prototype still it corresponds to (R7.1)	`node ui/scripts/screenshots.mjs` (extended for the new states), captures copied under `evidence/`
T6	Snapshot	n/a — DOM snapshots would freeze the markup this item removes and relocates; the role/name suite (T2) and the visual pass (T5) are the assertion
T7	Performance / load	n/a as a test — no data path changes; the bundle size is compared against 20.5.1 as the NFR check in T12 (removal must not grow it)
T8	Security / abuse case	yes	AC1: a fixture title `<b>x</b>`, a node name `n` and a transcript line with backticks render as text — no element created — through the one-line row, the row's `aria-label`, the header, the Session details popover and the trace; AC2: the popover's rows render from the typed `SessionRecord` / `PullRequestView` fields (the test passes a record whose fields carry markup and asserts `textContent`); fail-closed: the demo banner, the connection banner (`role="alert"`), the `degraded` health word, and the blocked composer's reason all render in the C layout; a grep for HTML-injecting calls is empty	`cd ui && bun run test`; `grep -rn "dangerouslySetInnerHTML\|innerHTML\|eval(" ui/src ui/index.html`
T9	Accessibility	yes	the search toggle and Full graph carry `aria-expanded`; Session details is a `<details>` whose summary is named; a sidebar row's accessible name carries `#n`, the title and the flag; `aria-current` on the selected row, tab and node; the trace keeps `role="log"`, its name and `tabindex`; the verbose switch answers Enter and Space; the `prefers-reduced-motion` guard is present; panels start closed below 1280 px (popover always) and 768 px (sidebar) and reopen from the header	`cd ui && bun run test`; `grep -n "prefers-reduced-motion" ui/src/styles/app.css`
T10	Migration / upgrade	n/a — no stored shape changes: the settings key, the theme pre-paint script and the hash grammar are untouched (R4.3, R5.4); T2 asserts every 20.5.1 hash lands on the same surface
T11	Manual exploratory	n/a — no live workstation service here; the demo transport exercises every verb, and the owner's review of the implementation PR, with T5's stills beside the chosen prototype's, is the human pass
T12	Lint / typecheck / build / bundle / untouched dirs	yes	the same three CI commands; the production bundle's JS and CSS sizes compared with the sizes at the branch point (JS 341 685 B, CSS 24 469 B — `bun run build` on 20.5.1); `git diff --stat <base> -- ui/src/api ui/src/state ui/src/demo` is empty	`cd ui && bun run lint && bun run build`; `ls -l ui/dist/assets`; `git diff --stat origin/main -- ui/src/api ui/src/state ui/src/demo`
T13	Markdown lint	yes	the spec chain, the decision record, `ui/README.md` and `docs/capabilities/control-plane.md` pass the repository linter	`npx --yes markdownlint-cli2@0.18.1 "*/.md"` from the repo root
T14	Documentation parity	yes	`ui/README.md` § Where the screens get their data and `docs/capabilities/control-plane.md` describe the C layout — what each band shows by default and what sits behind a disclosure — and no longer describe the session panel column or the graph strip band as current (R8)	review of the two files against the inventory; recorded in `evidence/documentation.md`

Scenarios & requirement trace

Row	Requirement(s)	Scenario / case
T1	R2.1, R2.5	`railNote` returns `parked — awaiting a human` for a parked item and the attention detail otherwise; never the node name
T1	R3.3	the sidebar's two-group mapping puts needs-you items first and everything else by last activity; `sidebarGroup` still returns the four-way answer for `itemStatus`
T2	R2.2, R3.2, R6.2	`Scenario: a sidebar row is one line carrying the number, the title and the urgent flag, and its accessible name carries all three`
T2	R2.2	`Scenario: a paused, merged or closed item shows no flag on its row`
T2	R3.2	`Scenario: pull request rows nest only under the selected item`
T2	R2.4	`Scenario: only the Needs you group shows a count`
T2	R3.3	`Scenario: the Standing group does not render in the sidebar and the Standing surface still lists every standing session`
T2	R3.3, R6.1	`Scenario: the search box and the instance filter appear behind the search toggle, which carries aria-expanded`
T2	R2.6	`Scenario: an instance filter renders on a fleet and not on a worker`
T2	R2.1, R3.2	`Scenario: the header meta is the ref chip, the phase chip, the note and the age — the repository line and the instance chip are gone`
T2	R2.1, R2.5	`Scenario: the current node is written once — the phase chip — and the gate banner says Human gate without the node`
T2	R3.3, R6.1	`Scenario: Full graph in the header meta reveals the node row and Collapse graph hides it`
T2	R2.6, R3.3	`Scenario: the session tab bar renders only when the item has more than one session`
T2	R2.4	`Scenario: the tab bar carries no session tally and the composer carries no delivery hint`
T2	R3.3, R3.4	`Scenario: pause, resume, stop and start render in the header and call controlSession with the work item's ref`
T2	R5.2	`Scenario: with a PR session selected the verbs carry a visible work item label and still target the work item`
T2	R3.3, R2.3, R6.1	`Scenario: Session details is a closed details element whose body shows harness, status, started, session id, transcript path, cwd, the attach command and Open on GitHub in full`
T2	R5.2	`Scenario: copy ref and copy attach command write the clipboard`
T2	R5.3	`Scenario: loading, empty board, missing deep link, no-transcript fallback, truncated tail and blocked composer render their sentences in the C layout`
T2	R5.4	`Scenario: every hash the 20.5.1 app parses lands on the same surface`
T2	R6.3	`Scenario: below 768 px the sidebar starts closed and reopens from the header`
T5	R1.1, R4, R7.1	the screenshot set in both themes, each still beside its prototype counterpart
T8	Sec AC1, AC2	the markup-bearing fixture through every relocated position; the popover from the typed record
T8	Sec fail-closed	the four warning signals present in the C layout
T9	R6.1, R6.2, R6.4	roles, names, `aria-expanded`, `aria-current`, `role="log"`, reduced motion
T12	NFR, R5.1	lint, types, build, bundle not larger, the three directories untouched
T13, T14	R8, NFR	markdown lint; the two documents describe the C layout

Verification environment

Repositories: this repo only, branch issue-511-declutter-control-plane.
Services / containers: none — demo mode. The screenshot script blocks nothing but needs no network; the Google Fonts request may fail offline, and the captures then render in system faces, which the evidence notes.
Fixtures & data: ui/src/demo/fixture.ts, unchanged (R5.1).
Browser: Chromium from Playwright's install (PLAYWRIGHT_BROWSERS_PATH) or CHROMIUM_PATH; Playwright itself is installed without saving (npm i --no-save playwright), as ui/scripts/screenshots.mjs documents — it is not a dependency of the app (NFR: no new dependency).
Credentials: none.
Bring-up: cd ui && bun install --frozen-lockfile && bun run build && bun run preview --port 4173. Tear-down: stop the preview server.
If bring-up fails: record it under Verification results, leave T5 unticked, escalate.

Evidence plan

Row	Evidence	Path under `evidence/`
T1, T2, T8, T9	vitest summary (files, tests, duration) and the list of new or changed test names	`unit.md` (T1), `integration.md` (T2, T9), `security.md` (T8)
T12	lint output; `tsc` + vite build output; the bundle sizes beside 20.5.1's; the empty diff of the three directories	`lint-and-typecheck.md`, `build.md`
T13	markdownlint output	`lint-and-typecheck.md` § Markdown
T5	one PNG per state per theme, indexed with the prototype still each matches; the before/after element count for the briefing (R7.2)	`ui/*.png`, table in `ui-visual.md`
T14	the two documents' sections quoted against the inventory	`documentation.md`

No flow needs an animated capture: every behaviour verified is a state (open/closed, on/off, selected) and a pair of stills shows it.

Verification activities

[x] T1/T2/T8/T9 — cd ui && bun run test
[x] T12 — cd ui && bun run lint && bun run build, then ls -l ui/dist/assets against the 20.5.1 sizes
[x] T12 — git diff --stat origin/main -- ui/src/api ui/src/state ui/src/demo (expect empty)
[x] T13 — npx --yes markdownlint-cli2@0.18.1 "**/*.md"
[x] T8 — grep -rn "dangerouslySetInnerHTML\|innerHTML\|eval(" ui/src ui/index.html (expect no matches)
[x] T9 — grep -n "prefers-reduced-motion" ui/src/styles/app.css (expect a match)
[x] T5 — node ui/scripts/screenshots.mjs <out-dir> against bun run preview over dist/, both themes, 1440×900 and 390×844; copy the captures under evidence/ui/
[x] T14 — read ui/README.md and docs/capabilities/control-plane.md against the inventory's C column; record in evidence/documentation.md

Verification results

Executed on 2026-10-20 on branch issue-511-declutter-control-plane at the head of the implementation commits, in demo mode, no credentials. Every planned activity ran and passed; the n/a rows stayed n/a as planned.

Activity	Command / procedure	Outcome	Evidence
T1 unit	`cd ui && bun run test`	pass — 22 files, 366 tests	`evidence/unit.md`
T2 integration · T9 accessibility	`cd ui && bun run test` (with `NODE_OPTIONS=--no-experimental-webstorage` on this Node 26 workstation; not needed on CI's Node 22)	pass — every scenario title of the trace table has a test	`evidence/integration.md`
T8 security / abuse case	`cd ui && bun run test`; `grep -rn "dangerouslySetInnerHTML\|innerHTML\|eval(" ui/src ui/index.html`	pass — abuse cases 1 and 2 and the four fail-closed signals; the grep has no match	`evidence/security.md`
T9 reduced motion	`grep -n "prefers-reduced-motion" ui/src/styles/app.css`	pass — one match, the guard unchanged	`evidence/integration.md`
T12 lint · typecheck · build	`cd ui && bun run lint && bun run build`	pass — oxlint clean; `tsc --noEmit` clean; built in 1.29 s	`evidence/lint-and-typecheck.md`, `evidence/build.md`
T12 bundle	`ls -l ui/dist/assets` against the branch point	pass — JS 337 157 B (was 341 685), CSS 24 421 B (was 24 469)	`evidence/build.md`
T12 untouched directories	`git diff --stat origin/main -- ui/src/api ui/src/state ui/src/demo`	pass — empty	`evidence/build.md`
T13 markdown lint	`npx --yes markdownlint-cli2@0.18.1 "*/.md"`	pass — 1764 files, 0 errors	`evidence/lint-and-typecheck.md`
T5 UI / visual	`node ui/scripts/screenshots.mjs <out> http://localhost:4173/the-loop/ui/` over `bun run preview --port 4173`	pass — 32 stills, both themes, 1440×900 and 390×844, no page error; before/after 56 → 27	`evidence/ui-visual.md`, `evidence/ui/`
T14 documentation parity	`ui/README.md` and `docs/capabilities/control-plane.md` read against the inventory's C column	pass	`evidence/documentation.md`

Re-run after self-review round 1 (SessionAside and the open-panel path removed, the composer's dead prop dropped): bun run lint, bun run test (22 files, 366 tests) and bun run build clean; JS 336 728 B, CSS 24 350 B.

Re-run after the critic round (evidence/critic-review.md: the empty-rail sentence, the popover's measured bound, the blocked pointer's aria-current, the verbs test): bun run lint, bun run test (22 files, 372 tests) and bun run build clean; JS 336 965 B, CSS 24 350 B; the 32 stills recaptured with no page error.

Not executed: T3, T4, T6, T7, T10, T11 — n/a as planned (no contract, data-path, snapshot, performance, migration or live-service change; the bundle check stands in for T7 and the owner's PR review, with the stills beside the prototype's, for T11).

Review comments

Appended by the-loop's record-feedback hook when a human gate approves with comments (issue-109). Append-only and attributed: an approval never silently discards a reviewer's suggestions, and the feedback travels with the document it concerns rather than living in a side-channel tracker.

2026-10-20 — approved

@MadaraUchiha-314 wrote:

approved