SolvNum

Release: Beta — suitable for trial and evaluation; APIs may change before GA

Deterministic arithmetic — bit-identical results on every platform

SolvNum is the deterministic-arithmetic substrate. It produces bit-identical numerical results across x86, ARM, NVIDIA GPU, and WebAssembly — the same trajectory, the same Greeks, the same attestation hash, every machine, every time.

Mechanism: bounded-update integer log-domain arithmetic with license-protected deterministic semantics. Sold as six vertical SKUs: Defense, Finance, ZK, Medical, Embedded, and Scientific — each licensed separately for its regulatory regime.

What it does

Floating-point arithmetic is not portable. The same C/Python code on x86, ARM, and GPU can produce slightly different results — fine for web apps, unacceptable when a regulator, validator, or certification authority requires reproducibility.

SolvNum is a deterministic number type: integer log-domain arithmetic with bounded-update semantics. The same inputs produce the same outputs and the same attestation hash on every platform. Use it as a substrate inside pricing engines, GN&C stacks, ZK witnesses, or anywhere you need to prove "this exact computation happened."

When to use it

Defense certification, SR 11-7 model risk, blockchain validator consensus, medical imaging reproducibility — sold as six vertical SKUs matching each regulatory regime.

See Examples for runnable code.

Who it is for

• Defense programs requiring bit-identical GN&C across heterogeneous hardware
• Finance teams under SR 11-7 / FRTB needing reproducible pricing kernels
• ZK/consensus protocols requiring validator-agreed arithmetic
• Medical imaging with FDA 510(k) reproducibility requirements

What ships

• C library (libsolvnum) — integer-only hot path
• Python wheel (solvnum) with SolvNum type
• Attestation hashing for cross-platform verification

Related products

On-chain variant: Blockchain Arithmetic. Pairs with SolvSRK for certified trajectories.