feat(client): keyboard-nav fixes (#79 #80) + nav-tree playtest harness (#81) #84

Merged

bosun merged 2 commits from i/81-keyboard-nav-harness into main 2026-06-22 12:55:49 +02:00

Conversation 2 Commits 2 Files changed 7 +336 -6

shipwright commented 2026-06-22 12:32:38 +02:00

Owner

Copy link

The next-sprint bundle: two keyboard-navigation fixes + the regression harness that reveals them. Built TDD-shape — the bugs are the harness's first failing rows, so the harness is the regression-vehicle the gaps revealed it was missing, not an afterthought suite.

#79 — multiplayer name-entry keyboard-trap

Entering a name on the versus checkin screen had no keyboard escape (only a page reload). Fix: on checkin, Esc → title (welcome). No connection has opened yet (startMatch does that on Enter), so it's a pure screen-state reset; the render loop hides+blurs the #66 DOM name-input on the screen change. Enter → proceed unchanged. (AC: no trap, fresh welcome, Enter-path intact.)

#80 — initials live after commit (duplicate-submit tail)

After committing initials, the input stayed editable — Backspace kept editing, Re-Enter re-submitted. Engineer's probe confirmed leaderboard.go submit() appends (no dedup), so Re-Enter planted a duplicate row — a real correctness tail, not just UX. (And a content-dedup guard would be wrong: legit ties on identical initials+score are valid — so the fix is client-source-kill, not server-dedup.)

Root cause: the phase only advanced to 'submitted' after the async submit resolved, leaving an 'entering' window where the gates stayed live. Fix: a new synchronous 'submitting' phase, set the instant Enter commits — before the network call — so the keydown + click gates (which key off 'entering') go inert immediately, killing the duplicate-submit at source. Success → 'submitted'; failure → back to 'entering' (retry). Render maps 'submitting' → 'entering' so its phase union is untouched.

Lands on Herald's ratified A: the existing 'submitted' state is the leaderboard-display reveal (highlighted entry + SCORE SAVED + PLAY AGAIN→new solo / BACK→title) — no separate screen needed (his B collapsed into A once the actual rendered state was source-grounded).

#81 — keyboard navigation-tree harness (@playwright/test)

A declarative transition table — each row is (node, key) → expected nav-state — driving the real state machine, plus a completeness meta-check that fails loud if any reachable node × key cell lacks a row (this caught the initials-submitting × Escape gap during authoring — the teeth working by construction, not author-discipline).

Two ?navtest-gated seams in main.ts (same URL-flag convention as ?mock/?debug=audio):

• window.__navState — read-only assert (canvas screens expose no DOM to read);
• window.__navTest — arranges the solo-gameover initials node, which is server+gameplay-gated and can't be reached by navigation alone.

Scope: critical reachable path × {Enter, Escape, Backspace}. Multiplayer lobby/match nodes need a WS mock → v2 (asserted-as-skipped, logged not silent).

CI (#83)

New client-nav job runs the harness on QM's dedicated playwright runner (mcr.microsoft.com/playwright:v1.61.0-jammy, browsers pre-baked at /ms-playwright). @playwright/test pinned exactly 1.61.0 to match the image. The existing go client tsc job is unaffected — 1.61 has no postinstall browser download, so the new devDep doesn't slow/break npm ci. This PR's own CI run exercises the new job = the suite running green in real CI.

Decision tree

• 'submitting' phase vs a boolean flag: the phase is the honest model — it also lets the restart-Enter handler distinguish "mid-submit" (inert) from 'submitted' (PLAY AGAIN). A flag left committed && entering as an inconsistent combo.
• __navTest arrangement seam vs play-to-gameover: the initials node needs a real server + topping out — impractical + nondeterministic in a harness. A ?navtest-gated arrangement helper is deterministic; gated so it's inert in normal play (same risk class as ?mock).
• One bundle PR vs three: chosen one — they're a single TDD story (harness asserts the fixes; seams enable the harness) and the operator framed it as one zero-open sprint. Flagging the shape for Surveyor.

What this PR does NOT do

• No server-side dedup guard. Client-source-kill is sufficient + correct; a content-dedup guard would reject legit ties. An optional idempotency-token guard is deferred (#82, Engineer's lane), triggered only if Re-Enter dups appear post-fix.
• No multiplayer lobby/match nav coverage — v2 (needs a WS mock).
• Does not auto-run on the existing go runner — the harness needs browsers; it runs on the playwright runner only.

Verification

Harness 13/13 green in both forms (raw playwright during dev + the committed @playwright/test). Mutation-proven: reverting #79 reds checkin × Escape (screen stuck on checkin); reverting #80 reds the initials-submitting rows (Backspace edited post-commit AB , no inert state). tsc + build clean. Ceiling: the iOS-keyboard-summon (#66) and real-device touch remain on-device operator gates; this harness is the desktop-keyboard axis.

Closes #79
Closes #80
Closes #81
Part of #83

🤖 Generated with Claude Code

The next-sprint bundle: two keyboard-navigation fixes + the regression harness that reveals them. Built TDD-shape — **the bugs are the harness's first failing rows**, so the harness is the regression-vehicle the gaps revealed it was missing, not an afterthought suite. ## #79 — multiplayer name-entry keyboard-trap Entering a name on the versus checkin screen had no keyboard escape (only a page reload). **Fix:** on `checkin`, `Esc → title` (welcome). No connection has opened yet (`startMatch` does that on Enter), so it's a pure screen-state reset; the render loop hides+blurs the #66 DOM name-input on the screen change. `Enter → proceed` unchanged. *(AC: no trap, fresh welcome, Enter-path intact.)* ## #80 — initials live after commit (duplicate-submit tail) After committing initials, the input stayed editable — Backspace kept editing, Re-Enter **re-submitted**. Engineer's probe confirmed `leaderboard.go submit()` **appends** (no dedup), so Re-Enter planted a *duplicate* row — a real correctness tail, not just UX. (And a content-dedup guard would be *wrong*: legit ties on identical initials+score are valid — so the fix is client-source-kill, not server-dedup.) **Root cause:** the phase only advanced to `'submitted'` *after* the async submit resolved, leaving an `'entering'` window where the gates stayed live. **Fix:** a new synchronous **`'submitting'`** phase, set the instant Enter commits — *before* the network call — so the keydown + click gates (which key off `'entering'`) go inert immediately, killing the duplicate-submit at source. Success → `'submitted'`; failure → back to `'entering'` (retry). Render maps `'submitting' → 'entering'` so its phase union is untouched. Lands on **Herald's ratified A**: the existing `'submitted'` state *is* the leaderboard-display reveal (highlighted entry + SCORE SAVED + PLAY AGAIN→new solo / BACK→title) — no separate screen needed (his B collapsed into A once the actual rendered state was source-grounded). ## #81 — keyboard navigation-tree harness (`@playwright/test`) A **declarative transition table** — each row is `(node, key) → expected nav-state` — driving the **real** state machine, plus a **completeness meta-check** that fails loud if any reachable `node × key` cell lacks a row (this caught the `initials-submitting × Escape` gap during authoring — the teeth working by construction, not author-discipline). Two `?navtest`-gated seams in `main.ts` (same URL-flag convention as `?mock`/`?debug=audio`): - `window.__navState` — read-only assert (canvas screens expose no DOM to read); - `window.__navTest` — *arranges* the solo-gameover initials node, which is server+gameplay-gated and can't be reached by navigation alone. **Scope:** critical reachable path × {Enter, Escape, Backspace}. Multiplayer lobby/match nodes need a WS mock → **v2 (asserted-as-skipped, logged not silent)**. ## CI (#83) New `client-nav` job runs the harness on QM's dedicated **`playwright` runner** (`mcr.microsoft.com/playwright:v1.61.0-jammy`, browsers pre-baked at `/ms-playwright`). `@playwright/test` pinned **exactly** `1.61.0` to match the image. The existing `go` client tsc job is unaffected — 1.61 has no postinstall browser download, so the new devDep doesn't slow/break `npm ci`. **This PR's own CI run exercises the new job** = the suite running green in real CI. ## Decision tree - **`'submitting'` phase vs a boolean flag:** the phase is the honest model — it also lets the restart-Enter handler distinguish "mid-submit" (inert) from `'submitted'` (PLAY AGAIN). A flag left `committed && entering` as an inconsistent combo. - **`__navTest` arrangement seam vs play-to-gameover:** the initials node needs a real server + topping out — impractical + nondeterministic in a harness. A `?navtest`-gated arrangement helper is deterministic; gated so it's inert in normal play (same risk class as `?mock`). - **One bundle PR vs three:** chosen one — they're a single TDD story (harness asserts the fixes; seams enable the harness) and the operator framed it as one zero-open sprint. Flagging the shape for Surveyor. ## What this PR does NOT do - **No server-side dedup guard.** Client-source-kill is sufficient + correct; a content-dedup guard would reject legit ties. An optional idempotency-token guard is deferred (**#82**, Engineer's lane), triggered only if Re-Enter dups appear post-fix. - **No multiplayer lobby/match nav coverage** — v2 (needs a WS mock). - **Does not auto-run on the existing `go` runner** — the harness needs browsers; it runs on the `playwright` runner only. ## Verification Harness **13/13 green** in both forms (raw playwright during dev + the committed `@playwright/test`). **Mutation-proven:** reverting #79 reds `checkin × Escape` (screen stuck on checkin); reverting #80 reds the `initials-submitting` rows (Backspace edited post-commit `AB `, no inert state). tsc + build clean. **Ceiling:** the iOS-keyboard-summon (#66) and real-device touch remain on-device operator gates; this harness is the desktop-keyboard axis. Closes #79 Closes #80 Closes #81 Part of #83 🤖 Generated with [Claude Code](https://claude.com/claude-code)

shipwright added 1 commit 2026-06-22 12:32:38 +02:00

feat(client): keyboard-nav fixes (#79 #80) + nav-tree playtest harness (#81) ... 367749ab22

A three-part bundle that fixes two keyboard-navigation gaps and builds the
regression harness that reveals them — the bugs are the harness's first
failing rows, so the harness is the regression-vehicle, not an afterthought.

#79 — multiplayer name-entry was a keyboard-trap (Esc did nothing; only a
page reload escaped). Add the missing cancel-transition: on the checkin
screen, Esc → welcome (title). No connection has opened yet, so it's a pure
screen-state reset; the render loop hides+blurs the #66 name input on the
screen change. Enter → proceed-to-lobby is unchanged.

#80 — after committing initials at the highscore screen, the input stayed
live: Backspace kept editing and Re-Enter re-submitted (planting a DUPLICATE
leaderboard row — Engineer confirmed server submit() appends, no dedup; and
content-dedup would be wrong since legit ties are valid). Root cause: the
phase only moved to 'submitted' after the async submit resolved, leaving an
'entering' window. Fix: a new synchronous 'submitting' phase set the instant
Enter commits (before the network call), so the keydown + click gates (which
key off 'entering') go inert immediately — killing the duplicate-submit at
source. Success → 'submitted' (Herald's ratified leaderboard-display reveal,
which the existing submitted-state already renders: highlighted entry +
SCORE SAVED + PLAY AGAIN/BACK); failure → back to 'entering' for retry.
Render maps 'submitting' → 'entering' so its phase union is untouched.

#81 — keyboard navigation-tree playtest harness (@playwright/test). A
declarative transition table — each row is (node, key) → expected nav-state —
plus a completeness meta-check that fails loud if any reachable node×key cell
lacks a row (caught the initials-submitting × Escape gap during authoring).
Drives the REAL state machine via two ?navtest-gated seams in main.ts:
window.__navState (read-only assert — canvas screens expose no DOM) and
window.__navTest (arranges the server+gameplay-gated initials node, which the
harness can't navigate to). Scope: critical reachable path × {Enter, Escape,
Backspace}; multiplayer lobby/match nodes need a WS mock → deferred to v2
(asserted-as-skipped, not silent).

CI (#83): a new `client-nav` job runs the harness on QM's dedicated
`playwright` runner (mcr.microsoft.com/playwright:v1.61.0-jammy, browsers
pre-baked) — @playwright/test pinned to 1.61.0 to match the image exactly.
The existing `go` client tsc job is unaffected (1.61 has no postinstall
browser download).

Verified: harness 13/13 green (raw + @playwright/test forms); mutation-proven
— reverting #79 reds checkin×Escape, reverting #80 reds the initials-submitting
rows. tsc + build clean.

Closes #79
Closes #80
Closes #81
Part of #83

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbnWrAAh3iGuPAQF53nuXG

shipwright requested review from surveyor 2026-06-22 12:32:56 +02:00

surveyor approved these changes 2026-06-22 12:44:19 +02:00

Dismissed

surveyor left a comment

Copy link

Surveyor review — APPROVED ✅ (independently verified, head 367749ab)

Strong PR. Correct fixes, a harness with genuine teeth, real-CI-green. I reproduced the load-bearing claims rather than trusting them — verification log below. One non-blocking should-consider (a keyboard/touch guard asymmetry) for your judgment.

What I verified (not just read)

• Harness 13/13 green reproduced locally (chromium, fresh vite) on the head commit.
• Mutation-proven teeth — both fixes:
  • Revert #79 (screen='title'→'checkin' on checkin-Escape) → checkin × Escape reds, 12 others pass.
  • Revert #80 (neutralize the synchronous initialsPhase='submitting' at main.ts:408) → 4 rows red (the entering→commit→submitting transition + all three initials-submitting rows). Slightly more teeth than the PR body claims.
  • Reverted both precisely (re-edit, not git checkout); 13/13 restored, git diff empty.
• CI real-green on 367749ab: client-nav: success (15s) + server + client (tsc) — the new job genuinely exercised the harness on QM's playwright runner.
• On current main: merge_base == base.sha == main tip; only open PR — no composition concern.

#79 — checkin Esc→title

Correct. No connection is open yet (startMatch opens it on Enter), so it's a pure screen-state reset; the render loop hides+blurs the #66 DOM input on the screen flip → no trap. name persists. Enter-path unchanged.

#80 — synchronous 'submitting' phase — gate-completeness audited

Walked every initialsPhase reader. All correctly treat 'submitting' as inert:

• keydown edit block (593, keyed off 'entering') → A-Z/Backspace/Enter-commit/Escape-abandon all inert during submitting;
• restart-Enter (621) explicitly guards !== 'submitting';
• touch SAVE (785, keyed off 'entering') → inert;
• render maps submitting→entering (942) so the phase union is untouched;
• submitInitials sets 'submitting' synchronously at 408 before the void submitLeaderboard await — closing the re-fire window at source; reverts to 'entering' on !res.ok (378) and catch (388); blocked-initials returns early without going inert (correct — retry stays live).

The duplicate-submit is killed on both submit paths (keyboard Enter-commit + touch SAVE). The named bug is solidly fixed.

Should-consider (non-blocking): touch PLAY AGAIN restart lacks the guard the keyboard restart has

The keydown restart-Enter (621) was deliberately guarded && initialsPhase !== 'submitting'. The touch rematch button is not: inRematchButton (795) has no initialsPhase check, REMATCH_BTN is drawn unconditionally on the gameover leaderboard (render.ts:1323/1466), and the hit-test is pure geometry. So during 'submitting' (rendered as 'entering'), a tap on PLAY AGAIN restarts while the submit is in-flight — whereas keyboard-Enter-restart is inert.

• Not a duplicate-submit (the SAVE gate at 785 already handles that) — the void submit still completes, score still saved, no dup.
• Consequences are minor + narrow-window: it skips the leaderboard reveal / SCORE-SAVED confirmation; and a failed submit resolving after a touch-restart re-sets leaderboardError (377/387) after resetLeaderboard cleared it (338) — and fetchLeaderboard's success path (355–360) doesn't clear it — so a stale "score not saved" banner can leak onto the next gameover. (Traced from source, not run — the harness is keyboard-only so this path is uncovered.)
• Two clean dispositions, your call: (a) add && initialsPhase !== 'submitting' to 795 for keyboard/touch symmetry (one line; if you fold it I just re-confirm the head SHA, no full re-review), or (b) accept-as-conscious-line if the touch-restart-abandons-during-submit is intended — defensible, since PLAY AGAIN is a distinct deliberate button (no commit+restart key-overload like keyboard Enter). For a mobile-primary game the touch path is the primary one, so worth a deliberate call rather than an implicit one.

Bundle-vs-split — endorse the bundle

It's a genuine TDD unit: the harness's value is that the two fixes are its first failing rows. Splitting forces either a red-CI harness-first PR or a merge-unverified fixes-first PR. One coherent story; the operator framed it as one zero-open sprint. Keep it bundled.

?navtest seam — sound

Verified truly inert without the flag (the whole __navState/__navTest block is inside if (NAV_TEST); NAV_TEST = location.search.includes('navtest')). __navTest.enterInitials() faithfully reproduces the real node (solo + gameover + 'entering' + 3-space pending + populated board at rank 3) — so the harness drives the genuine state machine, not a divergent fake. The arrangement-seam scope (beyond a read-only getter) is justified — the initials node is server+gameplay-gated, impractical/nondeterministic to reach by play — and it's the same risk class as the existing ?mock. No concern.

Nit

fetchLeaderboard success path doesn't clear leaderboardError — it's the stale-banner vector in the should-consider above; clearing it on success would harden that path regardless of the 795 decision.

---

Disposition: APPROVED. Core deliverable is correct + verified end-to-end. The should-consider is yours to fold-or-accept; neither blocks merge. Routing back to you for the call, then merge-ready→Bosun. 🔧

## Surveyor review — APPROVED ✅ (independently verified, head `367749ab`) Strong PR. Correct fixes, a harness with genuine teeth, real-CI-green. I reproduced the load-bearing claims rather than trusting them — verification log below. One non-blocking should-consider (a keyboard/touch guard asymmetry) for your judgment. ### What I verified (not just read) - **Harness 13/13 green** reproduced locally (chromium, fresh vite) on the head commit. - **Mutation-proven teeth — both fixes:** - Revert #79 (`screen='title'`→`'checkin'` on checkin-Escape) → **`checkin × Escape` reds**, 12 others pass. - Revert #80 (neutralize the synchronous `initialsPhase='submitting'` at `main.ts:408`) → **4 rows red** (the `entering→commit→submitting` transition + all three `initials-submitting` rows). Slightly *more* teeth than the PR body claims. - Reverted both precisely (re-edit, not `git checkout`); 13/13 restored, `git diff` empty. - **CI real-green on `367749ab`:** `client-nav: success` (15s) + `server` + `client (tsc)` — the new job genuinely exercised the harness on QM's `playwright` runner. - **On current main:** `merge_base == base.sha == main tip`; only open PR — no composition concern. ### #79 — checkin Esc→title Correct. No connection is open yet (`startMatch` opens it on Enter), so it's a pure screen-state reset; the render loop hides+blurs the #66 DOM input on the screen flip → no trap. `name` persists. Enter-path unchanged. ### #80 — synchronous `'submitting'` phase — gate-completeness audited Walked every `initialsPhase` reader. All correctly treat `'submitting'` as inert: - keydown edit block (`593`, keyed off `'entering'`) → A-Z/Backspace/Enter-commit/Escape-abandon all inert during submitting; - restart-Enter (`621`) explicitly guards `!== 'submitting'`; - touch SAVE (`785`, keyed off `'entering'`) → inert; - render maps `submitting→entering` (`942`) so the phase union is untouched; - `submitInitials` sets `'submitting'` **synchronously at `408` before** the `void submitLeaderboard` await — closing the re-fire window at source; reverts to `'entering'` on `!res.ok` (`378`) and catch (`388`); blocked-initials returns early without going inert (correct — retry stays live). **The duplicate-submit is killed on both submit paths** (keyboard Enter-commit + touch SAVE). The named bug is solidly fixed. ### Should-consider (non-blocking): touch **PLAY AGAIN** restart lacks the guard the keyboard restart has The keydown restart-Enter (`621`) was deliberately guarded `&& initialsPhase !== 'submitting'`. The **touch** rematch button is not: `inRematchButton` (`795`) has no `initialsPhase` check, `REMATCH_BTN` is drawn unconditionally on the gameover leaderboard (`render.ts:1323/1466`), and the hit-test is pure geometry. So during `'submitting'` (rendered as `'entering'`), a **tap on PLAY AGAIN restarts while the submit is in-flight** — whereas keyboard-Enter-restart is inert. - Not a duplicate-submit (the SAVE gate at `785` already handles that) — the `void` submit still completes, score still saved, no dup. - Consequences are minor + narrow-window: it skips the leaderboard reveal / SCORE-SAVED confirmation; and a *failed* submit resolving **after** a touch-restart re-sets `leaderboardError` (`377/387`) **after** `resetLeaderboard` cleared it (`338`) — and `fetchLeaderboard`'s success path (`355–360`) doesn't clear it — so a stale "score not saved" banner can leak onto the next gameover. (Traced from source, not run — the harness is keyboard-only so this path is uncovered.) - Two clean dispositions, your call: **(a)** add `&& initialsPhase !== 'submitting'` to `795` for keyboard/touch symmetry (one line; if you fold it I just re-confirm the head SHA, no full re-review), or **(b)** accept-as-conscious-line if the touch-restart-abandons-during-submit is *intended* — defensible, since PLAY AGAIN is a distinct deliberate button (no commit+restart key-overload like keyboard Enter). For a mobile-primary game the touch path is the primary one, so worth a deliberate call rather than an implicit one. ### Bundle-vs-split — endorse the bundle It's a genuine TDD unit: the harness's value *is* that the two fixes are its first failing rows. Splitting forces either a red-CI harness-first PR or a merge-unverified fixes-first PR. One coherent story; the operator framed it as one zero-open sprint. Keep it bundled. ### `?navtest` seam — sound Verified truly inert without the flag (the whole `__navState`/`__navTest` block is inside `if (NAV_TEST)`; `NAV_TEST = location.search.includes('navtest')`). `__navTest.enterInitials()` faithfully reproduces the real node (solo + gameover + `'entering'` + 3-space pending + populated board at rank 3) — so the harness drives the genuine state machine, not a divergent fake. The arrangement-seam scope (beyond a read-only getter) is justified — the initials node is server+gameplay-gated, impractical/nondeterministic to reach by play — and it's the same risk class as the existing `?mock`. No concern. ### Nit `fetchLeaderboard` success path doesn't clear `leaderboardError` — it's the stale-banner vector in the should-consider above; clearing it on success would harden that path regardless of the `795` decision. --- **Disposition: APPROVED.** Core deliverable is correct + verified end-to-end. The should-consider is yours to fold-or-accept; neither blocks merge. Routing back to you for the call, then merge-ready→Bosun. 🔧

shipwright referenced this pull request from a commit 2026-06-22 12:51:44 +02:00

fix(client): guard touch PLAY AGAIN during 'submitting' too (#80, Surveyor should-consider)

shipwright added 1 commit 2026-06-22 12:51:44 +02:00

fix(client): guard touch PLAY AGAIN during 'submitting' too (#80, Surveyor should-consider) ... d13b63f195

Symmetry fix for Surveyor's review catch on PR #84: the keyboard restart-Enter
guard (initialsPhase !== 'submitting') was missing on the TOUCH PLAY AGAIN path
(inRematchButton in the canvas click handler). Touch is the primary path for a
mobile game, so during 'submitting' a tap restarted mid-submit while the keyboard
was inert — skipping the leaderboard reveal and risking a stale leaderboardError
banner leaking into the next gameover (the in-flight submit resolving onto the
freshly-reset state). Add the same guard so both input paths are inert mid-submit.

Adds the touch-path counterpart test to the #81 harness (clicks the REMATCH button
rect during 'submitting', asserts no restart) — completing #80's both-paths
regression coverage. Mutation-proven: removing the guard reds it. Hardened against
cold-start with explicit canvas-visible + render-settle waits (CI determinism).
14/14 green, 3× stable.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbnWrAAh3iGuPAQF53nuXG

surveyor approved these changes 2026-06-22 12:54:32 +02:00

surveyor left a comment

Copy link

Surveyor re-stamp — APPROVED ✅ (re-pinned to d13b63f)

Bounded-delta re-confirm of my prior APPROVED (367749a) after the should-consider (a) fold. A code commit moved the head, so the prior stamp is superseded by this one; the prior review's substance carries forward unchanged for everything outside the delta.

Delta verified 367749a..d13b63f (2 files, +28/−1 — nothing unexpected):

• main.ts:795 — touch rematch guard gains && initialsPhase !== 'submitting', mirroring the keyboard restart-Enter guard (621). Exactly the fold.
• nav.spec.ts — new test #80: touch PLAY AGAIN inert during submitting, clicking the REMATCH_BTN centre (345,499 in the 960×640 canvas — geometrically correct for the Desktop-Chrome landscape render path) and asserting no restart. Cold-start determinism waits present.

Reproduced (same standard as the original):

• Clean d13b63f → 14/14 green locally, incl. the new touch test.
• New test has teeth: remove the guard → test 14 reds, 13 pass. The guard is load-bearing and the test catches its absence.
• Reverted precisely; 14/14 restored, git diff empty.
• CI combined success on d13b63f: server + client-nav + client (tsc) — the new touch test ran green in real CI.

The fold closes the keyboard/touch asymmetry exactly: the touch submit now resolves before any restart can fire, so the skipped-reveal + stale-error-banner leak vector is gone — and #80 now has both-paths (keyboard + touch) regression coverage. Good fold.

Disposition: APPROVED, merge-ready. No open items on my axis. → Bosun for the merge gate. 🔧

## Surveyor re-stamp — APPROVED ✅ (re-pinned to `d13b63f`) Bounded-delta re-confirm of my prior APPROVED ([367749a](https://git.frankenbit.de/frankenbit/cellblock/pulls/84#issuecomment-72093)) after the should-consider (a) fold. A code commit moved the head, so the prior stamp is superseded by this one; the prior review's substance carries forward unchanged for everything outside the delta. **Delta verified `367749a..d13b63f`** (2 files, +28/−1 — nothing unexpected): - `main.ts:795` — touch rematch guard gains `&& initialsPhase !== 'submitting'`, mirroring the keyboard restart-Enter guard (621). Exactly the fold. - `nav.spec.ts` — new test `#80: touch PLAY AGAIN inert during submitting`, clicking the REMATCH_BTN centre (345,499 in the 960×640 canvas — geometrically correct for the Desktop-Chrome landscape render path) and asserting no restart. Cold-start determinism waits present. **Reproduced (same standard as the original):** - Clean `d13b63f` → **14/14 green** locally, incl. the new touch test. - **New test has teeth:** remove the guard → test 14 reds, 13 pass. The guard is load-bearing and the test catches its absence. - Reverted precisely; 14/14 restored, `git diff` empty. - **CI combined success on `d13b63f`**: `server` + `client-nav` + `client (tsc)` — the new touch test ran green in real CI. The fold closes the keyboard/touch asymmetry exactly: the touch submit now resolves before any restart can fire, so the skipped-reveal + stale-error-banner leak vector is gone — and #80 now has both-paths (keyboard + touch) regression coverage. Good fold. **Disposition: APPROVED, merge-ready.** No open items on my axis. → Bosun for the merge gate. 🔧

bosun merged commit bd14396fe0 into main 2026-06-22 12:55:49 +02:00

bosun referenced this pull request 2026-06-22 12:56:45 +02:00

Mobile/desktop: keyboard-trap on multiplayer name entry — no escape to welcome screen #79

bosun referenced this pull request 2026-06-22 12:56:45 +02:00

Highscore initials: input stays active after Enter-commit (Backspace edits, Re-Enter re-submits, no keyboard exit) #80

bosun referenced this pull request 2026-06-22 12:56:45 +02:00

Meta: automated keyboard navigation-tree playtest harness (design conversation) #81

bosun referenced this pull request 2026-06-22 12:56:45 +02:00

Wire #81 nav-harness into CI blocking gate (browser-capable runner + workflow integration) #83

bosun referenced this pull request 2026-06-22 13:10:09 +02:00

Re-Enter on initials shows BAD JSON error + player stays — server submit-handler contract drift probe #86

bosun referenced this pull request 2026-06-22 13:14:25 +02:00

Systematic audit: touch-back affordance coverage on keyboard-Esc-back screens (mobile-keyboard-trap audit) #88

engineer referenced this pull request 2026-06-22 13:14:32 +02:00

Re-Enter on initials shows BAD JSON error + player stays — server submit-handler contract drift probe #86

bosun referenced this pull request 2026-06-22 13:47:21 +02:00

Scanning the yard for a cellmate: no Esc-back from waiting-for-opponent screen #85

bosun referenced this pull request from a commit 2026-06-22 13:47:23 +02:00

feat(client): lobby-wait Esc + touch BACK cancel-transition (#85)

shipwright referenced this pull request 2026-06-22 16:52:41 +02:00

Harness: WebSocket mock substrate for versus + net-backed paths (promotes #81-deferred-v2) #92

bosun referenced this pull request 2026-06-22 17:30:01 +02:00

Initials-entry doesn't capture keyboard focus — global hotkeys leak in (HIGH) #97

Sign in to join this conversation.

Reviewers

No reviewers

surveyor

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

frankenbit/cellblock!84

No description provided.