Engineering Standards
How we build matters as much as what we build.
Test Coverage
Internal testing covers cryptographic correctness, GPU computation, governance rules, synthetic benchmarks and integration. The prototype supports bounded information sharing and signed decision records under the stated trust model.
FCNS: Falsifiable Detection
The engine includes the Financial Crime Network Simulator (FCNS): a synthetic benchmark that measures AML detection accuracy against realistic criminal behaviour. Eight FATF typologies. Three jurisdiction packs (NL, UK, DE) calibrated against national statistics. Adversarial criminals that adapt when caught. Systemic noise from real banking infrastructure quirks. Macro-economic shocks. 16 formally defined scoring metrics. Every detection claim is testable, reproducible, and falsifiable. No other vendor can prove their detection works. We publish the proof.
GPU-Native Computation
500K entities under 2 seconds on the fixed synthetic fixture, with about 55K transactions. This is a detection test, not the full product cycle.
Privacy-Preserving Federation
The prototype supports bounded information sharing and signed decision records under the stated trust model.
Defence-Grade Build
The engine is compiled with memory protection (buffer overflow detection, address space randomisation, data execution prevention), control flow integrity (preventing return-oriented programming attacks), side-channel mitigations (Spectre variant protection), and runtime memory corruption detection. These protections are enabled for every build configuration, not just release builds. The security posture exceeds typical financial services software requirements and aligns with standards expected in critical national infrastructure.
Reproducible Binaries
Every compilation of the same source code produces a byte-identical binary. No timestamps, no random seeds, no build environment leakage. A regulator or auditor can independently verify that the deployed binary matches the audited source code. Supply-chain integrity is a build property, not an afterthought.
Maximum Strictness
The codebase compiles at the highest available warning level with warnings treated as errors. Static analysis runs on every build. The engine enforces C++23 standard conformance with no compiler-specific extensions. Build-time contract enforcement ensures that GPU compute kernels and CPU data structures remain aligned. ABI mismatches between components are detected at compile time, not at runtime.
Proven Cryptography
All cryptographic operations use published, peer-reviewed algorithms. Ed25519 for digital signatures. X25519 for key agreement. Elliptic curve Diffie-Hellman for oblivious transfer base operations. Blake3 for hashing with domain separation. No custom cryptography. No proprietary ciphers. Every implementation is validated against known-answer test vectors from the relevant specification documents.
Exact Arithmetic
Financial computations use fixed-point arithmetic with explicit precision guarantees. No floating-point rounding errors in risk score comparisons or threshold evaluations. GPU compute kernels enforce the same precision constraints as CPU reference implementations, verified by cross-validation tests.
Entities under 2 seconds in a fixed synthetic test, about 55K transactions. Detection test only.