Visual Tests
This public proof gallery is designed around one rule: a buyer should understand the point of a scene in three seconds and trust it in thirty. Each test now defaults to a cleaner Public proof view, adds an Event focus lane for the key moment, and keeps a stronger Debug proof mode for technical inspection.
Proof at a glance
Buyer-critical launch path
What you are seeing
Real UPE capture loops with visible slabs, terrain or tank walls, contact-friendly framing, and subsystem-specific cues for suspension, cloth pins, SPH waterlines, and benchmark lanes.
Why it matters
These scenes reveal solver instability, replay divergence, or weak integration claims quickly instead of hiding them behind dense docs or presentation-only polish.
Pass signal
Look for planted contact, obvious scene intent, 3–5 readable telemetry numbers, and the ability to switch between Public proof, Event focus, and Debug proof without losing the point of the scene.
- Box Stack Stability — Stable stack under repeated runs with consistent contact resolution.
- Mixed Shape Resting Contact — Mixed primitive contacts settle cleanly without obvious interpenetration drift.
- Friction Ramp Comparison — Material tuning produces readable, repeatable sliding behavior.
- CCD Tunnel Prevention — Fast-moving bodies preserve contact intent under CCD-enabled conditions.
- Snapshot / Restore Parity — Snapshot restore preserves the expected replay path in the reference scene.
- Same Seed Same Result — Same seed, same binary, same output path under deterministic mode.
- Vehicle Suspension Course — Vehicle dynamics are part of the sale story, not an afterthought add-on.
- Canonical Benchmark Sweep — Performance claims are tied to canonical scenes and stored artifacts, not a single vanity number.