9.3 KiB
9.3 KiB
LANE HANDOVER LOG
Coordination file between lanes (Opus 4.8 execution sessions) and the reviewer (Fable). Append-only — never edit or delete previous blocks. Newest block at the BOTTOM. Push after every session so the reviewer can see state remotely.
Protocol
- Before starting work: read this whole file + your lane file in
lanes/+docs/CONTRACTS.md. Check for a REVIEW block addressed to your lane — it overrides your lane file where they conflict. - Work only inside your lane's owned directories (see BUILD_PLAN.md §2).
- End of session: append a SESSION block (template below), commit, push your
lane/<name>branch. - Never merge to main — the reviewer does that.
- Blocked on a contract ambiguity? Don't stall and don't silently change the
contract: stub locally with
// TODO(contract):, log a CONTRACT CHANGE REQUEST below, keep moving.
SESSION block template
---
### SESSION — <LANE-NAME> — <date> <time>
**Branch/commits:** lane/<name> @ <short-sha>..<short-sha>
**Done:** bullet list of what actually works now (be honest — "renders but untested" ≠ "done")
**How to see it:** exact command + what to click (e.g. `npm run dev` → press D → deny 3 patrons)
**Tests:** X passing / Y failing (name the failures)
**Broke / known-wonky:** anything you left imperfect
**Contract change requests:** none | numbered list with justification
**Questions for reviewer:** none | numbered list
**Next session should:** bullet list
REVIEW block template (reviewer only)
---
### REVIEW — Fable — <date>
**Reviewed:** <lanes/commits>
**Merged to main:** <yes/no + sha>
**Contract decisions:** rulings on any open change requests
**Instructions for <LANE>:** what to do next / what to fix first
(log starts below)
SESSION — LANE-0 (executed by Fable, reviewer) — 2026-07-19 16:30
Branch/commits: main (reviewer executed Phase 0 directly; no lane branch) Done:
- Vite + strict-TS + Phaser 3.90 + Vitest + ESLint scaffold;
Math.randombanned by lint rule. - All CONTRACTS.md types verbatim in
src/data/types.ts. core/: SeededRNG (mulberry32, named streams), typed EventBus (snapshot-emit semantics), GameClock (real-ms driven,ageOnbirthday math incl. leap-day), Meters (single-writer, hype multiplies positive vibe only, clicker drift), StubBeatClock (128 BPM), versioned corrupt-safe localStorage save.data/: outfit vocabulary (+PALETTE), 9 archetypes, 10 contraband items, 17 Dazza texts (8 rule-bearing with ruleIds).patrons/: generator (fake-ID tell system, DOBs clustered near the 18 boundary, late-arrival drunk drift), puredollPlan+ PhaserrenderDoll(3 poses, drunk tells baked), regulars' memory store.rules/drunk.ts: drunkStage thresholds.- ParadeScene demo: rainy street, kebab-shop + venue dressing, 24-walker parade with ID cards rendered from the same data, SPACE reseed / D toggle cards.
- 49 tests green; lint/build/test gate clean.
How to see it:
npm run dev→ http://localhost:5199 (in the Claude preview pane, drive frames per CLAUDE.md "Dev notes"). Watch for analmost18label walking under an ID card claiming 18+ — that's the whole game in one image. Tests: 49 passing / 0 failing. Broke / known-wonky: - ID cards overlap when the footpath is crowded — cosmetic, parade-only.
window.gamedebug handle exposed in main.ts (deliberate, for pane-driving).- Spec deviations: doll renderer split into
dollPlan.ts(pure) +doll.ts(Phaser) — better than spec'd, plans are pixel-testable without a GPU; added a 5th fake-ID tell (expired real card, no boolean tells); no MainMenu scene (Boot → Parade direct) — menu is Phase 2 shell work. Contract change requests: none (contracts implemented as written). Questions for reviewer: n/a (self-reviewed). Next session should: Phase 1 lanes are OPEN — see REVIEW block below.
REVIEW — Fable — 2026-07-19
Reviewed: Phase 0 (self-executed; verified via test gate + in-browser parade). Merged to main: yes (direct). Contract decisions: CONTRACTS.md now frozen. Extension requests via this file. Instructions for Phase 1 (LANE-DOOR, LANE-FLOOR, LANE-JUICE):
- Branch from current main into
lane/door/lane/floor/lane/juice(worktrees if sharing a machine). Read your lane file's new "Phase-0 reality" section FIRST — it corrects the original spec where the landed code differs. - LANE-DOOR is the critical path; if only one Opus session is available, run DOOR first, JUICE second, FLOOR third.
- Every session ends with the gate + a SESSION block here + a push. Fable reviews on request ("review the lanes").
SESSION — LANE-JUICE — 2026-07-19 18:55
Branch/commits: lane/juice @ c262ac8..2eb3c94 Done:
audio/TechnoEngine.ts— synthesized 128 BPM track (kick / off-beat hats / 2-bar 16th bassline / 8-bar pad), look-ahead scheduler (25ms timer, 100ms horizon) as the beat authority. Emitsbeat:tick {beatIndex, audioTimeMs}with the scheduled time, plusbeat:baron downbeats. Drop-in replacement forcore/StubBeatClock. Music → lowpass → master; SFX bypass the filter.audio/Sfx.ts— 10 procedural one-shots +startRain/stopRain. One shared noise buffer; voices disconnect onended.audio/scheduling.ts+audio/filterCurve.ts— pure, node-testable beat grid and cutoff-ramp math (the WebAudio classes are thin wrappers over these).ui/—Stamp,Phone,DialogueBox,Clicker,MeterHud, plusstyle.ts(chunky widget kit) andtypo.ts(deterministic drunk typos).ui/JuiceDemoScene.ts— exercises every widget and SFX; log-mapped cutoff slider, DOOR/FLOOR toggle routed through the bus, beat pad that flashes whenctx.currentTimereaches the scheduled time (not on event receipt). How to see it:npm run dev→ http://localhost:5199/#juice (or press J; P returns to the parade). Click once to unlock audio, then: drag MUFFLE to sweep the filter by hand, hit STEP INSIDE for the door→floor sweep, mash the SOUNDS column, stamp someone with NOT TONIGHT. Tests: 111 passing / 0 failing (was 49 at Phase 0; +62 this lane). Verified by ear: NOTHING. No audio output available in this environment. Everything below was logic-checked in a real browser via the exposed engine:- door→floor sweep ramps 250Hz→18kHz over 600ms, monotonic, exponential (t=0.5 lands ~2.1kHz vs the ~9.1kHz a linear ramp gives).
- under a genuinely hidden/throttled tab: 52 consecutive beats, zero
scheduled in the past, 75–98ms lead, max 1 beat scheduled per timer wake.
John/Fable still need to ear-check: does the kick read as a kick, is the
bassline musical, is
typeTickquiet enough at the real mix level (it fires per character), and does the sweep feel like a door rather than a filter. Broke / known-wonky: src/main.tsis edited (2 lines + a ~12-line demo route). Outside LANE-JUICE ownership — see request 2 below. Without it the demo scene is unreachable and Vite tree-shakes it out entirely.TechnoEnginehas no test file — it needs a realAudioContextand the suite runs in node. Its scheduling/filter math is tested via the pure modules, but theresyncarithmetic is only reachable through the class. ExtractingresyncBeat(grid, nextBeat, nowMs)intoscheduling.tswould close that gap.Phonethread truncates an over-tall newest message rather than scrolling.Stampdecals are caller-owned now; the demo caps them at 5. The door scene must destroy them (or passparent) or they accumulate over a night. Contract change requests:
- Add
door:phoneTheatre: { durationMs: number }toEventMap. Emitted byui/Phonewhen the player closes the phone, carrying how long they stared at it while a queue waited — the door scene rewards it with hype. It is a door event, not an audio one, so it is outside my namespace. I did not editdata/types.ts.ui/juiceEvents.tsdeclaration-merges it intoEventMapinstead: fully typed, zero edits to the frozen file, and folding it into the real contract later is a copy-paste. Same file also addsaudio:unlockedandbeat:bar, which ARE mine per LANE_JUICE.md but are declared together so the whole juice surface reads in one place. - Sign-off requested on the
src/main.tsedit (scene registration + J/#juice route, marked// TODO(integration)). Happy to revert it if you would rather wire the demo from the Phase 2 shell yourself. Questions for reviewer: TechnoEngine.unlock()anchors beat 0 atcurrentTime + 60msso beat 0 is as honest as every other beat. If a consumer assumes beat 0 lands exactly ataudio:unlocked.atMs, that is a 60ms discrepancy — is that acceptable, or shouldatMscarry the anchored origin instead?MeterHudcounts heat strikes fromheat:strikeevents it receives. On a mid-run scene swap it would restart from 0. Should it take an initial count in its options, or readNightState.heatStrikes.lengthat construction? Next session should:
- LANE-DOOR: swap the local stamp for
ui/Stamp(// TODO(juice)), and takeMeterHudin place ofHudStub. - Someone with speakers should ear-check the mix and tell me what to retune.
- Consider extracting
resyncBeatintoscheduling.tsso the throttled-tab resync gets direct test coverage.