# Road-edge response capture (r352, 2026-10-16)

## Viewer adjustment

The combined-city viewer sampled the center and four recovered wheel points
for road support, but required three compatible samples even when the center
was supported. On Route 736 the center and one wheel remained on the authored
road while the other wheel points fell beyond the pavement mesh. Movement was
rejected and the car stopped on the inside edge of the curve. The support rule
now accepts a road-supported center plus one compatible wheel; when the center
is exactly on a mesh seam, the previous three-wheel fallback remains.

Rejected movement also used a fixed `0.35` speed multiplier once per physics
substep of at most 4.4 ms. This made a road edge act as an almost instant-stop
wall. The viewer now applies its existing 12 m/s² preview braking rate over
the substep duration. These are viewer collision choices, not recovered
retail curb or tire behavior.

## Packaged drive capture

On a fresh Route 736 load, a 3 s W hold reached 7.22 m/s; 1.5 s of A reached
11.06 m/s. Steering back with D reached the inside edge and the car remained
stationary once its center left the road support. Under the time-scaled
response, speed decayed instead of disappearing in one frame. S then backed
the car about 5 m away from the edge to `−7.08 m/s` (reverse), with no browser
page or console errors. The inspected image is
[`506-r352-edge-brake-reverse-recovery.png`](../webgl-city/review/images/506-r352-edge-brake-reverse-recovery.png).

The final frame still shows the car close to the curb; this check does not
show that the vehicle can recover from every unsupported position. It confirms
only this road-edge case and does not calibrate retail steering or braking.
The visible gray body paint also remains an unresolved preview material.