[Testing][E2E] Route-affordance walk: doc precision and dead-branch follow-ups from the #2679 review

[Chris0Jeky]

Follow-ups from the fresh-context review of PR #2679 (head 2df5d4c39), all non-blocking under the severity bar; none is a correctness defect in a shipped row. Test-only, the three files PR #2673 created.

1. Header pointer (MEDIUM in the review, doc precision). tests/e2e/route-affordances.spec.ts:36 says which defence applies to each walked response row is "stated per row in the inventory". The header's own two-bullet enumeration is accurate for all ten walked response rows, but home.quick-capture-submit, inbox.archive and review.approve carry no comment in routeAffordanceInventory.ts, and calendar.next-month's comment explains only the exact-string post-condition. Either add the per-row note to those four rows or drop the "stated per row" clause.
2. Token-substitution doc names the wrong surfaces (LOW). routeAffordanceInventory.ts:72-75 says the tokens are substituted in url and response pattern strings; {openingMonthLabel} appears only in text.text and textChangedFrom.from, and fillTokens is applied to value.value, text.text, textChangedFrom.from and role-selector name, not to attribute.*, enabled.selector, node/focus test ids or css selectors. State the real list, or apply fillTokens uniformly.
3. assertEnabled reads are dead branches (LOW). The guarded status variant declares assertEnabled: true as a required literal, so the guard's !== true throw and the walk's if (affordance.status.assertEnabled) cannot take their false branch, and the assertion-8 docblock describes a per-row choice the type forecloses; unlike the response branch (assertion 9), the check has no canary. Either widen the type to boolean with a canary, or say in both places that the type is the enforcement and the runtime read is belt-and-braces.
4. textChangedFrom is a negated matcher (LOW, hypothesis to confirm). expect(locator).not.toHaveText(x) is expected to pass when the locator resolves to zero elements. Assertion 8 classifies the kind as element-asserted and offers it for guarded rows. The shipped calendar.previous-month row is masked by calendar.next-month's positive text check one row later. Confirm the zero-element semantics; if confirmed, pair the kind with an explicit toHaveCount(1) (or toBeVisible()) on the same locator before the negation.
5. The calendar mount read is armed without the parked-outcome handler (LOW). route-affordances.spec.ts:542-547 arms the calendar GET before page.goto and awaits it on the next line; if page.goto throws, the 30 s waitForResponse rejection has no handler. The metrics, notifications and home mount reads predate PR test(e2e): exhaustive consequence dispatch, calendar mount consumption and truthful post-condition docs in the route walk #2679 and have the same shape. Apply the parking pattern from activate to all four.

Also noted, informational: a postCondition: { kind: 'response' } on any row passes typecheck and the vitest guard and throws only inside E2E Smoke (expectConsequence's response case); that is the documented design, recorded here so the limit of the cheap gate is written down.

Refs #1949, #2678, PR #2673, PR #2679.

[Chris0Jeky]

[Claude lane claim v2]
lane: alpha-product-trust (session taskdeck-alpha-85, Taskdeck-Alpha checkout)
issue: #2682
base: origin/main at claim time (f74514f or later)
residual acceptance: the five follow-ups from PR #2679's review, one test-only PR on the three files PR #2673 created: (1) the header pointer at route-affordances.spec.ts ~36 becomes true by adding the per-row defence note to the four rows that lack one (home.quick-capture-submit, inbox.archive, review.approve, and calendar.next-month beyond its exact-string post-condition) rather than by dropping the clause; (2) the token-substitution docblock at routeAffordanceInventory.ts ~72-75 states the real substitution list (value.value, text.text, textChangedFrom.from, role-selector name) and where {openingMonthLabel} actually appears, instead of widening fillTokens to selectors and attributes; (3) assertEnabled: the guard's !== true throw and the walk's runtime read stay as belt-and-braces, and both places plus the assertion-8 docblock say that the required-literal type is the enforcement and the runtime read cannot take its false branch, unless the worker finds a cheaper canary that actually proves the branch; (4) textChangedFrom: confirm from Playwright's own source in node_modules whether expect(locator).not.toHaveText(x) passes on zero elements; if it does, pair the kind with an explicit toHaveCount(1) on the same locator before the negation (in expectConsequence, not per row) and say so in the assertion-8 docblock, so calendar.previous-month is no longer masked by the next row's positive check; (5) the calendar, metrics, notifications and home mount reads armed before page.goto adopt the parked-outcome pattern activate already uses, so a goto failure leaves no unhandled 30 s waitForResponse rejection. The informational note (a postCondition: { kind: 'response' } on any row throws only inside E2E Smoke) stays recorded on the issue, not changed. Proof: the vitest inventory guard spec green (Test Files count equal to the paths named); the E2E route-affordances spec run against a running stack (the Gemini env vars unset per the box's rule) with the walked rows green, or an explicit NOT-verified line naming what blocked the run; typecheck on the frontend; eslint on the changed files; git diff --check.
owned paths/modules: frontend/taskdeck-web/tests/e2e/route-affordances.spec.ts; tests/e2e/routeAffordanceInventory.ts; the inventory guard vitest spec PR #2673 created. NOT owned: app source under src/ (a doc-precision PR must not change product behaviour), the other E2E specs, the Playwright config
shared-path leases: none; no open PR touches the three files (checked gh pr list files at claim time)
parallel-safe work: #1940's PR #2705 (Paper review rail) is disjoint
open PRs/branches/review threads checked: both claim forms on #2682 (none; the issue has no comments; filed by the lane from PR #2679's review on 2026-09-05); PRs #2673, #2679 merged
human or decision gates: none
status: active (worktree codex-2682-route-walk-docs, being created)

[Chris0Jeky]

Follow-up recorded from PR #2706's review (not bundled there): the four mount reads parked by readOnMount have no event timeout of their own under this config (actionTimeout defaults to 0 and nothing sets one), so a mount read that never gets a response hangs to the test timeout (90 s in the board-seeded walk, 45 s in the Home/Inbox block) before reporting, the shape activate already bounds with { timeout: 15_000 } to protect the shared E2E Smoke budget. Pre-existing (the inline waits were equally unbounded). Decision: pass a bound to the mount reads matching activate, after checking CI's single-worker lane timings for the first read after web-server start; one line in readOnMount plus its docblock.