moat G — same seed + params → identical replay, anywhere
Run A (seed)
5e1c3431
x=340.6207 · v=81.6682 · steps=1000
Run B (same seed)
5e1c3431
x=340.6207 · v=81.6682 · steps=1000
Run C (seed + 1)
eae21194
x=340.6310 · v=81.6688 · steps=1000
A and B use the same seed and must produce identical state hashes. C uses a different seed and must differ — proving the seed actually drives the run.
No-Fibs — what this proves
A deterministic, seeded physics kernel means any simulation result is reproducible: hand someone the seed + params + step count and they get the identical run. That turns a physics claim into a verifiable artifact (No-Fibs for physics) and is the foundation for deterministic netcode and ghost replays. Honesty: cross-device floating-point determinism holds while you stay on the CPU and use Math.fround; GPU shaders and arbitrary-precision libs can break it.