When the stones let go.
Boulders tip, logs tumble and fragments scatter as the crossing loses its support. The routes between players stay open.
The first falling crossing changed the battlefield correctly, but its rocks moved like passengers on one slab. This pass separates their motion: each stone tips and falls around its own centre while the ground drops away beneath it.
Weight and separation
The large rocks and logs now have separate pivots, short differences in release timing, inward drift and accelerating rotation. The terrain slab tilts and falls away ahead of most of its dressing. Small stone fragments and restrained dust make the break easier to read without covering the whole event.
The motion is authored and deterministic. It is not a rock-to-rock physics simulation, and the falling dressing cannot damage creatures or obstruct the surviving routes. Two quieter rumble cues follow the initial break in the game; the attached video is silent.
There is still a way around
The five-second warning, trigger and instant of ground-collision removal are unchanged. Only one of four crossings is lost. Ground troops can still reach both altars and every font by another pass, and flying creatures can cross the gap.
Souls remain recoverable on the bank. Pause freezes the independent pieces; restart restores their transforms, visibility and original collision settings.
Checks and limits
The source game and final Mac app pass 99 checks on the Passes and 67 on the open island. New cases cover separate falling stones, non-colliding dressing, pause, exact restoration, repeatability and temporary debris cleanup. Native viewport input passes all 19 checks.
The first new collision assertion caught a prop retaining an active collision mode. Explicit non-colliding profiles fixed it, while preserving each original profile and response for reset. The earlier playable app and saved map remain unchanged.
The crossing still has angular authored edges. This iteration improves the falling motion; human presentation acceptance and the previously identified frame-rate problem remain open. A separate short real-time sample at 1280 × 720 High reported 27.0 ms average frames during the warning and 31.6 ms during the fall, with a 400 ms reported hitch. The fixed-step video does not show live frame-rate behaviour; that hitch needs follow-up.