Immersive native entry, environmental response and measured performance

[hoainho]

Closes #130

Scope

• Replace the always-running TitleStage with an actual 2560×1440 native Greenwake plate and lightweight CSS motion; preserve English controls and the candidate banner.
• Make introduction/story text and Continue/Skip immediately readable; retain real subtitle settings, keyboard/gamepad handler and live OS Reduced Motion. Loading exposes actual phases/asset progress, not a decorative countdown.
• Diagnose the bridge screenshot: its figure was baked into /asset-source/ui/maps/greenwake.jpg, used as a CSS background, not a live actor penetration. Replace that backdrop with a native scenery-only plate without altering CameraDirector/support authority.
• Add native house scuffs/bounded flex and two native loose wooden crate stacks with swept motion, settling and break; rooted trees remain fixed. Simulation owns damage; presentation never changes quest/reward state.
• Bake the existing blocked-candidate simplification recipe into five heavy native dependencies, compact vertices and resize embedded textures. Preserve native bindings/skins/selected vertex values and leave asset qualification blocked. Add a conservative path-sweep broadphase preserving narrow-phase decisions/order.
• Integrate on merged feat: remake native enemy and boss attacks #132 (a380076); retain native attack queue/sync/presentation and gate-birth timing.

Observed acceptance

• Production UI: keyboard title activation → readable intro → Escape skip → phased loading → real dialogue/next line/Skip Cinematic → tutorial dismissal and movement.
• Held required Yara GLB for 14 seconds: 54%, Loaded 41 of 49 scene assets, no gameplay HUD; release reached the actual dialogue.
• Responsive 390×844 and 960×540: no horizontal overflow. Live OS Reduced Motion: zero running title animations; mounted intro static with zero particles. Synthetic gamepad A/B samples exercised the actual handler (not physical-controller certification).
• Actual GameApplication/native scene + real K/J contacts: house health 180→148.07 without translation/collapse; tree 137.50→130.30 without translation; crate 80→57.40→0, removed colliders, no campaign-fact change during isolated break. Hero placement was seeded near each material before input; no camera/damage override.
• 84 native catalog output hashes match. Five native derivatives match the actual existing client recipe's ordered triangle attributes and fitted bounds. 2,672 exact path comparisons + six mutable collider/support transitions: zero mismatches.

Matched production measurements

One sequential cold/warm pair, one owned headless Brave/Chromium155 on AppleM1/16GB, 1568×817/DPR1. Same Aren/Koro Greenwake UI route, 2s hero-preview and1s intro dwell. CPU/frame, sampled CPU, all-pass GL and GPU query windows measured separately. Baseline is merged #131 main a380076, not the earlier pre-#131 run.

Metric	Before	After
Cold / warm observed title ready	1770 / 2114ms	878 / 745ms
Selection→first dialogue, cold / warm	24.19 /23.88s	8.26 /8.15s
Cold title transfer	20.12MB	1.52MB
Encoded resources through cold gameplay	489.59MB	142.42MB
Game JS / counter window	5173 /5187ms	1693 /5013ms
Game RAF p95 / long tasks	116.7ms /43	16.7ms /0
Post-GC game JS used / backing storage	24.68 /112.03MB	22.97 /62.41MB
Separate GPU median / p95	32.06 /34.28ms	19.73 /20.74ms
All-pass draws / triangles	80 /1,158,061	84 /1,160,805

Query-only differential runtime 2412→215ms is not an FPS benchmark. Draws/triangles increase slightly for native crates; no batching-reduction claim. Three normal title/new-run cycles retain stable game backing storage after change, but title used heap rises13.44→14.38MB and DOM329→345→361 (baseline also +16 nodes/cycle): not leak-free certification. One host/pair is not fleet percentile, locked60fps, all-boss stress or production asset acceptance.

Verification

• TypeScript + Vite build: pass,331 modules,8.62s wall; existing >500KB warning remains.
• Focused UI/environment/geometry/native candidate gate: 88 pass,0 fail,38,037 assertions,8 files.
• Node24 release tests: 236 pass,0 fail.
• Prior unrelated full-suite hero timeouts and Aren animation failure remain disclosed; no full-suite pass claimed and no repeat solely to confirm them.
• Production pageErrors empty. Favicon404 and navigation-aborted audio requests disclosed; audio playback not accepted.

Evidence

Acceptance ledger · UI/performance documentation · bridge/native budget diagnosis

Portable actual H.264/yuv420p/24fps/1280×720/fast-start videos:

• Entry UI — 37.1s
• Native crate/house/tree contacts — 20.1s

No paid generation, commercial binary decompilation, dependency/lockfile change or unrelated worktree edit. Commercial Game1/2 references are presentation references only. Temporary probes and unused derivatives removed; worktrees preserved. Main session owns the two-stage spec/code review; no independent external approval is claimed.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
wild-crown	Ready	Preview	Oct 18, 2026 4:29pm UTC

[hoainho]

Main-session review of 1ec0d1d. Coordinator review, not independent external approval.

1. Spec compliance: all #130 acceptance items exercised and mapped in acceptance.json. Actual native pre-rendered title; immediately readable introduction/dialogue; truthful held-required-asset loading; source-based bridge backdrop diagnosis; real native house/tree/crate offensive response; fresh matched production measurements against merged #131 main; English, candidate banner, keyboard/gamepad handler, live Reduced Motion and camera/simulation ownership preserved. No paid generation or commercial binary/source claim.

2. Code quality/state lifetime/regression: TitleStage and every old callback/caller removed; native preview lifetime remains separate. Narrative/choice text escaped. Versioned baked candidate marker only skips the same expensive client recipe for blocked candidates; hashes and native clip/skin validation remain fail-closed. High-water typed sweep buffers retain indices, not departed region objects; conservative AABBs preserve original order/narrow phase, and no recursive path query occurs within the sweep. Rooted structures never translate/collapse/change support. Broken crates stop receiving hits, sleep with zero velocity and lose their colliders. Actual fixed-step offense—not VFX—owns environmental damage. #131 cue queue/native sync/presentation ordering and current-frame gate birth retained.

Observed gates:88 focused tests,38,037 assertions;236 release tests;TypeScript+Vite build;84 catalog hashes;5 exact native-client geometry equivalence checks;2672 path queries+6 mutable transitions,zero mismatches. Actual production entry/readiness and native contact videos reviewed. No blocking finding identified within this scope.

Residual limits are intentional and published: native candidates remain blocked; not full-suite/all-boss/release or locked60fps approval; prior unrelated hero test failures disclosed; +16 DOM nodes per title cycle on both baseline/current prevents leak-free claim; audio playback not accepted; synthetic gamepad samples are not physical hardware. Await both build runs and deployment checks before merge.