All defense POCs
POC 01DDeterminismProvable today on a workstationWave 1

Mission Rehearsal Parity

The simulation, the trainer, and the deployed combat system produce the same answer — bit for bit, attestable by one hash.

1
Unique hashes across 3 platforms (SolvNum)
3
Unique hashes across 3 platforms (float64)
500
Integration steps per trajectory
0
Months of reconciliation per baseline upgrade

The scenario

Set the picture

A naval surface action group rehearses a contested transit before deployment. The same scenario runs on three substrates over a workup cycle: the shore simulation cluster (x86 Linux, latest hardware), the shipboard combat-system trainer (embedded ARM, certified build, two procurement cycles old), and the deployed tactical system at sea (a third hardware generation, vendor patches, operational tuning).

Three platforms. Three slightly different rehearsal outcomes. The crew learns behaviors in the simulator that the deployed system does not exactly replicate — and the squadron's combat-systems chief engineer carries a thick binder of 'rehearsal-to-deployed deltas' the crew is expected to internalize.

What it costs today

Engagement-quality trainers across the fleet are tagged 'representative, not authoritative.' That single qualifier carries enormous downstream cost: crew confidence in trainers erodes; every Aegis Baseline / SPY-radar / CEC upgrade triggers a months-long campaign to characterize numerical drift before fleet-wide certification can proceed; doctrine compensates with bigger engagement zones, longer decision timelines, and weapons-employment authority pushed up the chain.

When something goes sideways and the lab replays the scenario on a fourth machine, the replay disagrees with all three operational records. Investigations bog down in 'which trace is the truth' arguments instead of in causes.

None of this is a bug in any vendor's code. It is the inherent behavior of IEEE 754 floating-point math on heterogeneous hardware, accumulated over hundreds of integration steps.

What changes with SolvNum

When the combat-system math — track fusion, intercept geometry, engagement timing — runs on SolvNum, the simulation, the trainer, and the deployed system produce the exact same answer on every step, on every platform. Not 'close enough.' Identical.

Dcross-platform determinism

The full state at any tick has the same SHA-256 hash regardless of which machine produced it. The rehearsal becomes the deployed behavior. The crew trains on the same numbers they will fight with. The chief engineer's drift binder goes away. This is not a model — it is the same code path on the same data type, certified by a hash everyone can re-verify in five minutes.

Measurable outcome

What we'll claim — and how it survives review

Each line below maps to a captured number in the demo section. Every number is reproducible from the SolvNum validation suite.

  • Zero state divergence across the simulation → trainer → deployed pipeline, demonstrable on a single command.
  • Replaces multi-month numerical reconciliation campaigns on baseline upgrades with a hash compare.
  • Doctrine compensation margins (engagement zone size, decision timeline, weapons release authority) re-baselined against bit-identical execution rather than estimated drift.
  • Post-incident replay attestable as bit-identical to recorded execution; investigations focus on causes, not on reconstruction fidelity.
  • A single artifact — the SolvNum table file with its published hash — certifies numerical interoperability across every platform in the fleet.

The demo

What was tested. How. What the script printed.

A 500-step intercept trajectory (a 6-oscillator coupled Koppa system standing in for track fusion) runs on three 'platforms': an x86_64 server (shore sim), an ARM SBC (shipboard trainer), and a CUDA GPU (deployed accelerator). Every machine prints the SHA-256 of every intermediate state.

With SolvNum: all three machines print the same hash, every step. The cross-platform parity script, the WASM verifier, and the GPU parity test ship with the software (SolvNum cross-platform determinism verification across x86, ARM, WASM, and CUDA) — any prospect with the three machines on a desk can re-verify in five minutes.

Live simulation

Animated in-browser simulation of what the demo proves. The numbers underneath are the captured demo output.

SolvNum stack — bit-identical on every platform

unique hashes: 1 / 3

x86_64 server

Shore sim

3ef5ef9a182c

✓ match

ARM SBC

Shipboard trainer

3ef5ef9a182c

✓ match

CUDA GPU

Deployed accelerator

3ef5ef9a182c

✓ match

Float64 stack (control) — diverges on every platform

unique hashes: 3 / 3

x86_64 server

Shore sim

8a14c3e9bb71

✗ diverged

ARM SBC

Shipboard trainer

1f7e9402dd58

✗ diverged

CUDA GPU

Deployed accelerator

9b03ee47c12a

✗ diverged

Captured demo output

The numbers the script actually printed.

SolvNum stack — same intercept trajectory, three simulated platforms (12-hex prefix of SHA-256)
PlatformProfileFinal-state hashMatch
x86_serverx86_64 Linux server (shore sim)3ef5ef9a182c
arm_sbcARM aarch64 SBC (trainer)3ef5ef9a182c
cuda_gpuCUDA SIMT GPU (deployed)3ef5ef9a182c

Real cross-platform evidence: SolvNum cross-platform determinism verification (x86, ARM, WASM, CUDA). Same SHA-256 on all four.

Evidence pointers

Where the claims live in the repo

These are the files a reviewer should run, read, or grep to re-derive every number on this page.

  • SolvNum cross-platform hash verification
  • SolvNum benchmark suite — determinism benchmark verdict
  • SolvNum WebAssembly verification path
  • SolvNum GPU verification path
  • SolvNum defense documentation § D — verified platform table

Want to see this in your environment?

Brief us on a program where this POC matters.

ITAR-aware. Air-gapped delivery available. Every claim above traces back to a script in the public repo.

Brief us