Lock the polish branch's behaviour so a future UI edit surfaces as a failing
assertion to re-agree or fix.
Unit (vitest, node env):
- placement: recallIndex, cellOccupied/isBlankSlot, non-linear direction, the
single-tile submit default, and placementFromHint blank-fallback / rack-exhausted.
- banner: the marquee scroll-cycle repeat-then-advance, stop(), root-relative and
multiple links.
- client.GatewayError. Extract the check-word constraints out of Game.svelte into a
pure lib/checkword.ts (sanitize + canCheck) and cover them.
E2E (playwright mock, Chromium + WebKit):
- commit via the 🏁 control, history slide-down + close, the exchange dialog,
check-word input sanitising + verdict, resign-to-finished, and the Settings
board-label mode changing the on-board labels.
Add a webkit project to the Playwright config so the hermetic mock-mode specs run
in both Chromium and WebKit, and install both browsers in CI. WebKit's Debian build
runs headless without extra host system libraries (verified locally: smoke + zoom
pass in webkit); the workflow comment records the one-time host install-deps fallback
if a runner ever lacks a library. Desktop WebKit does not reproduce iOS Safari's text
auto-inflation, so the app.css text-size-adjust guard stays outside e2e coverage.
The Playwright MCP writes page snapshots/screenshots into a .playwright-mcp/
directory while inspecting the UI; keep that scratch output out of the tree.
Drive the focus-centring with requestAnimationFrame across the ~0.25s width
transition instead of a single scrollTo after transitionend. The board now stays
locked on the placed cell as it grows, removing the visible 'centre top-left, then
correct' double motion.
- AdBanner: move the side inset onto the scrolling track so the long message
scrolls to its very end; pin body text-size-adjust:100% so iOS/Safari stops
inflating the long marquee text.
- Game: do not zoom on drag start (the player may change their mind) — zoom and
centre happen on drop, in attemptPlace; a stray tap on an occupied cell no
longer cancels the rack selection (wait for an empty cell).
- Board: centre the focus cell after the zoom width transition finishes (was
clamping to top-left mid-transition); compute the cell from the rendered
scrollWidth.
- item 5: move container-type to the zoom-scaled .scaler so cqw labels grow WITH the board (magnifying-glass zoom); new e2e measures the font grows ~1.85x
- item 8: confirm popovers anchor to the trigger's right edge (no longer run off-screen)
- item 9: last-word flash runs 2 cycles then settles to normal (was infinite)
- nav bar grows ONLY in game (other screens: minimal nav, content fills); tab bar always bottom
- tab bar: tighter icon/label spacing, bigger icons, hint badge on the icon corner
- board zoom reworked to width-based (real native scroll, fixes Safari/Chrome) + constant cqw labels; pinch & swipe-to-history dropped (conflict), double-tap kept, history via menu
- beginner bonus labels shrunk to fit cells
- Draw opens exchange directly (no confirm); confirm popovers restyled like the hamburger dropdown (vertical); removed the floating direction toggle
- pending tiles darker bg (no outline); last-word dark-tile highlight (static / 1s flash)
- check button disabled for <2/>15 chars, already-checked, or 5s cooldown
- global user-select:none (inputs exempt); docs updated; TODO-4 alphabet-on-wire
System libs are now provisioned once on the runner host (install-deps); the job
downloads the browser into the runner cache and runs the smoke strictly (no
--with-deps, no sudo, no continue-on-error; timeouts kept as a hang guard).
The Gitea runner is an act host executor without apt privileges, so
'playwright install --with-deps chromium' (apt-get) hung ~51m then failed. Drop
--with-deps, bound the browser-install + e2e steps with timeouts and mark them
continue-on-error so a runner lacking GUI libs can't block the gate. The strict
gate stays check/unit/build/size; the e2e smoke remains a hard local pre-push check.
Engine: multi-player drop-out-and-continue with a per-game tile disposition (remove default / return), resigned seats skipped and excluded from the win, leaver rack never revealed; 2-player behaviour unchanged.
New domains (service/store, no HTTP yet): internal/social (friend request/accept graph, per-user blocks, per-game chat with nudge as a message kind, content filter via mvdan.cc/xurls/v2 + leet/separator normaliser + phone heuristic) and internal/lobby (in-memory variant-keyed matchmaking pool, friend-game invitations invite->accept with lazy 7-day expiry). account gains profile editing and the email confirm-code flow (Mailer seam: SMTP or log mailer).
Migration 00003_social.sql + regenerated jet. main wires the new services into the server (accessors for the Stage 6 handlers); robot substitution stays in Stage 5, REST/stream/push in Stage 6/8. Docs (PLAN, ARCHITECTURE, FUNCTIONAL+ru, TESTING, README) updated.
internal/game drives the engine over a single match and owns everything the
engine does not: event-sourced persistence (a games row + an append-only decoded
move journal, with the live engine.Game kept warm in a cache and rebuilt by
replay on a miss), the play/pass/exchange/resign transitions with
validate-at-submit scoring, an unlimited score/legality preview, the hint
(per-game allowance + profile wallet), the word-check tool with complaint
capture, per-player game state, history and GCG export (Poslfit dialect), and a
background turn-timeout sweeper that auto-resigns overdue turns honouring each
player's daily away window. Like Stages 1-2 it is a service/store layer with no
HTTP; the gateway surface lands in Stage 6.
Engine: additive decoded domain API (Direction, SubmitPlay/SubmitExchange/
EvaluatePlay/HintView/Hand, MoveRecord.{Dir,MainRow,MainCol}, Registry.Lookup,
ParseVariant) so internal/game never imports scrabble-solver; and a Resign fix
so the resigner keeps their score yet never wins (the other player wins a
two-player game). Timeout reuses Resign.
Persistence: migration 00002 adds games, game_players, game_moves, complaints,
account_stats and extends accounts with away_start/away_end/hint_balance; go-jet
regenerated. account gained SpendHint. Config adds BACKEND_DICT_DIR (required),
BACKEND_DICT_VERSION, BACKEND_GAME_TIMEOUT_SWEEP_INTERVAL, BACKEND_GAME_CACHE_TTL;
main loads the registry at boot (hard dependency) and starts the sweeper.
Tests: engine resign + decoded-API tests; game unit tests (GCG, away-window
boundaries, hint budget, cache, keyed mutex, payload); inttest integration
(lifecycle, replay equivalence, timeout sweep with away grace, resign stats,
hint policy, word-check/complaint, per-game-lock). Docs (PLAN, ARCHITECTURE,
FUNCTIONAL +_ru, TESTING, README) updated.
backend/internal/engine wraps the sibling scrabble-solver library in-process:
- Registry: versioned DAWG load via dafsa.Load, keyed by (variant, dict_version),
latest-per-variant; English / Russian / Эрудит handled uniformly.
- Bag: own deterministic, seeded tile bag with Draw + Return (for exchanges),
since the solver's self-play bag cannot return tiles.
- Game: pure rules engine — deal, play/pass/exchange/resign, refill, per-move
scoring, turn order, and end-condition detection (empty bag + empty rack, six
scoreless turns, resignation) with end-game rack adjustment.
- decode/ReplayBoard: dictionary-independent MoveRecords and board replay via
scrabble.Apply (no internal/encoding), realising ARCHITECTURE §9.1.
Wiring: go.work gains "replace scrabble-solver => ../scrabble-solver"; backend
requires scrabble-solver (placeholder) and github.com/iliadenisov/dafsa directly.
Both Go CI workflows clone the public solver sibling (master HEAD, no token) and
set BACKEND_DICT_DIR.
Docs: ARCHITECTURE §5/§14, TESTING engine layer, backend README, and PLAN
refinements + deferred TODOs (publish/version solver; split engine vs dictionary
generator).
Ilia Denisov
developer