Deterministic · Auditable · Forkable simulation — bit-identical on every machine.
Berk DT isn't a physics engine — it's a decision-verification infrastructure: it forks a live state to evaluate possible futures, fingerprints every step, and proves that the same input yields a bit-identicalresult on every machine.
A simulation used to make a decision has one question: can you trust it?On floating-point engines, the answer is fragile.
The same scenario yields different results on a different CPU/compiler. The prediction drifts on its own; you can't tell whether a divergence is model error, a fault, or cyber-tampering — so you loosen the threshold and go blind. "The simulation ran" is a claim you can't prove.
All math is fixed-point integer → the simulation side has zero-bit noise. The same input, the same state_hashon every machine. You compare the result in one line, reproduce it in one command, and catch tampering instantly.
This claim is auto-verified by CI on three platforms on every change; see it live in the Hash Consensusand CPU ≡ GPU ≡ browserdemos.
A classic digital twin monitors and replays (what happened?). Berk DT goes one step further: it forks futuresfrom a live state (what else could have happened?).
Each branch runs under a distinct plan; the result state is scored and hashed. Same base + same plan → same result. The options (COAs) you produce are defensible and reproducible; the base state is never mutated.
A deterministic twin runs beside the real system on the same inputs, tracking the residual between prediction and measurement and raising an alarm on the first frame of divergence. With zero simulation noise the threshold can be tight — a small wear/fault/injection trips it cleanly.
Advisor and Shadow Twin are the operational layer of the Sentinel edition; they share the same core as rollback netcode and the proof chain.
Decision safety, operational optimization, and sim-to-real verification — all from a single deterministic core.
Determinism is now the goal of many engines; the difference is howit's achieved and what it comes with. Berk DT unites in one core the three things competitors hold separately.
Strong physics, but cross-platform determinism depends on float + compiler flags (fragile) and there's no decision/twin layer.
Fixed-point and deterministic, but Unity only + its own cloud; game-netcode focused, not a digital twin.
Berk DTfills the intersection: by-construction fixed-point determinism + an embedded decision/twin layer + formal verification + an operational layer (Shadow Twin · Advisor) — a single lightweight, embeddable core.
We don't say "trust us." Thanks to determinism, every claim is reproducible in one command.
# Determinism: the golden hash is bit-identical on every platform cargo test --release --test determinism_dynamics # Forensic Inspector: generate a run, then re-verify it by PHYSICS cargo run --bin berk-audit -- run cargo run --bin berk-audit -- verify report.json # → all 600 frames reproduced ✓ # 10,000 bodies: SAP and BVH broadphase produce the SAME hash cargo test --release --test scale_10k -- --ignored --nocapture
Note: the interactive demos are currently labeled in Turkish, but they are visual and language-light — the hash badges, controls, and effects read the same in any language.