Berk Digital Twin — provable simulation infrastructure
A deterministic decision / digital-twin core with a built-in cross-platform physics engine. Same setup + same fixed timestep = bit-identical state_hash on every platform — the foundation of lockstep, replay and auditable autonomy.
Fork the moment. Prove the future.
Most engines run slightly differently on every machine. Berk Digital Twin is deterministic: fixed-point math + logical clock + seeded RNG → Windows ≡ Linux ≡ macOS ≡ browser, bit-identical state_hash.
Replay & recording
Seed + ordered inputs + periodic state hashes reproduce a run exactly. Incident analysis, debugging, training — divergence is caught at the first frame.
What-If engine
Fork N branches from live state, simulate each forward, score and hash-address them. Generated options are defensible and reproducible (the Advisor layer).
Shadow Twin
A deterministic twin runs alongside the real system on identical inputs, tracking the prediction–measurement residual. Zero simulation noise → even the smallest fault or injection breaks cleanly.
Rollback netcode
GGPO-style input prediction + rewind over snapshots: smooth, low-latency lockstep multiplayer even on bad networks — Photon Quantum parity, embedded in the core.
Explainable decisions
Every decision carries its epistemic state, time layer (reflex → strategic), authority and a provenance hash. "Why was this decided?" has a recorded, reproducible answer.
Engine-ready
C-ABI + Unreal plugin + Unity (UPM) package + WebAssembly. The game engine provides visuals; deterministic physics, replay and explainable decisions come from this core.
Game-engine-class, production-ready physics
Fixed-point (Fx64 Q32.32) 2D + 3D rigid-body dynamics with the TGS-Soft solver — proven allocation-free on the hot path.