Make your AI-generated code robust.
AI agents write more of our code every month — fast, plausible, and unproven against what actually breaks software: race conditions, lost writes, wrong results, and runaway memory. We build open-source tools that surface those failures before your users do — deterministically and reproducibly, as plain libraries in your own cargo test. Two are live.
$ cargo test --test catches_planted_bug ── DST FAILURE ───────────────────────── seed: 7 invariant: acked_writes_are_durable broke @ 12 minimal: CrashRestart @ step 12 ← from 40 re-run: SEED=7 cargo test catches_planted_bug ──────────────────────────────────────── $ SEED=7 cargo test --test catches_planted_bug # reproduces the exact failure, every time
It runs your real code — in your own test.
This is the open-source, developer-first way to bring deterministic simulation testing to a Rust component you own: inject a few seams, run it in your normal cargo test, on your laptop. No platform, no rewrite, no separate agent — and it exercises the code you actually ship, not a model of it.
A library, not a platform
You don't run your system inside a VM or under a separate tool — it's a dependency in your test binary. Start with one module, on a laptop, for free.
Scoped by design — and honest about it
Determinism reaches exactly as far as the seams you route through, for the Rust code you own. Whole-system, multi-language, escape-proof testing needs a dedicated platform — this stays focused on Rust, in your own test, as the fastest way to get most of the value first.
Established methods, one family, developer-first.
Each library corresponds to an established testing approach — not a term we invented — packaged as a focused crate, with a command-line toolchain to adopt them. Start with what's live; the rest are being extracted from what we already run against Navian Pulse.
Reproduce concurrency and crash-recovery bugs from a seed and shrink each failure to a minimal failing trace — the deterministic-simulation-testing approach, as a Rust library.
scan · migrate · explain · check — find determinism leaks (tiered, SARIF/CI-ready), auto-wire the safe time seams, and gate CI. The fastest way to adopt navian-dst.
Assert state stays bounded and memory growth levels off after warmup across long runs — leak and unbounded-growth detection under load, as a cargo-test assertion.
soak — run any command under load and gate CI on the plateau property: RSS must level off after warmup, or the build fails. No test code, no allocator to install.
Run your system and a small reference model under the same inputs and faults, then diff their state — catches silent data loss and divergence.
Verify relations between outputs when there's no obvious oracle — shift-invariance, monotonicity, scale-equivariance. Ideal for statistics and ML code.
Categories grounded in the standard taxonomy (deterministic simulation testing, differential/model-based testing, metamorphic testing, resource-leak testing). See the roadmap →
These tools were built while hardening Navian Pulse — a real-time decision engine with a sub-millisecond core and replayability as a design requirement. The reliability work is the same; the open-source family is the part that generalizes to any codebase.
Read our story →Two tools are live. Start with one module.
Add a deterministic test, or a memory-plateau assertion, to a single module — in your own cargo test, on a laptop, for free.