Made of little things.

# Impact Yard verification

Verified on 2026-09-14. Preview: <http://127.0.0.1:4184/> through Napplet Space,
with the built artifact inside a real `sandbox="allow-scripts"` iframe.

## Delivered

- Procedural Three.js yard, car, seven destructible structures, and a ramp.
- 157 independent cannon-es rigid bodies, four-wheel raycast suspension,
  120 Hz fixed physics steps, friction and collision impulses.
- Keyboard and pointer driving, three cameras, orbit drag, brake, slow motion,
  pause, car recovery, yard rebuild, and impact/displacement telemetry.
- Optional theme and storage SDK integration; no hard domains or relay calls.
- Product surfaces: `index.html`, `src/main.ts`, `src/simulation.ts`,
  `src/styles.css`, local metadata, dependencies, usage docs and regression tests.
  Boilerplate build substrate and guidance guards retained.

## Checks

| Check | Result |
| --- | --- |
| `pnpm type-check` | Pass |
| `pnpm build` / Napplet Space build watcher | Pass; one bundled `dist/index.html` |
| `pnpm test:guidance` | 6 passed |
| `pnpm test:physics` | 7 passed |
| `pnpm test:conformance` | 5 passed, 0 failed, 5 skipped |
| `napplet-space check --json` | Pass, profile `space-playback-1` |
| `node tests/browser-smoke.mjs` | Pass, zero browser errors |

Conformance skips signed manifest checks because no signed event was supplied,
wire checks because its domain-absence scenario emitted no envelopes, and
lifecycle checks because teardown evidence was not supplied. It does verify boot,
sandboxing, forbidden globals and graceful domain absence. The separate host
browser test exercises a user-triggered camera preference write/read via SDK
storage and sends the documented theme change through the installed runtime.

The browser test checked 200×160, 320×560, 900×600 and 2400×1200 in **both dark
and light themes**. No document overflow or clipped controls; frame fills the
available space. Root backgrounds follow the theme. There is no title header or
minimum-size gate. The game and camera controls also work without any injected
optional domains. Pointer accelerator press/release was tested at 320×560.

Physics checks establish that all 157 pieces remain stable for eight simulated
seconds at rest, impacts collapse the main stack without disturbing remote
structures, steering works in both directions, brakes decelerate, reverse moves
backward, and rebuild/recovery restore their intended state. The new tests drive
the car up the ramp from a standing start at both 20% and full throttle, collapse
the timber tower and loaded arch by driving into their supports, and confirm
that falling domino panels continue displacing their neighbors after braking.

## Ramp regression

The old ramp was a tilted 40 cm thick slab with a roughly 31 cm entrance lip.
At 20% throttle the car stalled near z=2.28 and reached only 0.65 m chassis
height. A diagnostic run lowering only the slab by 31 cm cleared the blockage
without changing the car, tire grip, slope or engine. The final fix replaces
the slab with a wedge whose top begins at ground level; its thickness is below
ground at the entrance. Renderer and collider share the same vertices/faces.
The regression test failed before this change and passes afterward.

New obstacles are a 14-panel domino run with rubber feet, a 27-piece timber tower
with alternating layers, and an 11-piece masonry arch/lintel/crate assembly.
All are freestanding; no timed toppling animation or artificial impact forces
were added. The structure guide shows seven entries and scrolls in shorter
frames. Optional domains, hard requirements and relay behavior are unchanged.

The built artifact scan found no direct network/browser-storage APIs and no
external scripts or styles. Rendering assets and libraries are bundled.

Screenshots and machine-readable browser results are in
`/tmp/impact-yard-checks/`. These temporary artifacts can be regenerated with
`pnpm test:browser` while the preview runs.

## Limits

This is a rigid-body approximation: it does not model material fracture,
splintering, tire deformation or vehicle body deformation. Headless Chromium
measured approximately 20 rendered frames/second at 1440 pixels wide on this
test machine; GPU/device performance has not been benchmarked. Physics timing
is fixed, with capped catch-up to avoid a backlog on slow devices.

No publication was performed. Signed-manifest and measured teardown conformance
remain outside the supplied local test evidence.