01
Behavioral coverage
Line coverage says which code ran. Behavioral coverage says which states, transitions and trajectories the system reached. Ordinum optimizes for the second.
Product, section 01
Fuzzing for the behavior of autonomous systems. Not just their code.
Ordinum searches the space of faults, timings and states your tests never reach, then hands every failure back as a minimized, replayable regression test.
01 /Evidence, not dashboards
THE PRODUCT, section 001
Failures you can hand to an engineer.
A failure is only useful if someone can reproduce it, understand it and keep it fixed. Every part of Ordinum is built around that.
- 02
Stateful fault injection
Faults are sequenced in time against a running system: a stale reading at tick 4, a retry at tick 9. Order and timing are part of the input.
- 03
Deterministic replay
Every run is seeded and recorded. A failure replays identically, every time.
- 04
Failure minimization
Delta debugging strips every fault that is not required to trigger the violation. What remains is the cause, not the noise.
- 05
Regression generation
Each minimized failure is emitted as a regression test. The next change that reintroduces it fails before it ships.
- 06
Cross-component verification
Invariants are checked across the whole loop, from perception to actuation. The unit under test is the system.
02 /A different question
WHY ORDINUM, section 002
Everyone else answers a different question.
Existing tools are good at what they were built for. Most check the cases someone wrote down. Few search for the interactions between components that nobody anticipated.
Traditional testing
Does it do what we expected?
- Finds
- Regressions in the cases you wrote down.
- Misses
- Only as broad as the engineer's imagination. Interactions between components go untested.
Security tools, pentesting, red teaming
Can an attacker break in?
- Finds
- Exploitable vulnerabilities and adversarial inputs.
- Misses
- Aimed at attackers and malicious input, not at timing, retries and stale state in a control loop.
Observability
What went wrong out there?
- Finds
- What already happened in production.
- Misses
- Tells you after the incident. For a robot, that is after the collision.
Ordinum
What can this system do that nobody anticipated?
- Finds
- Unexpected behavior across components, before deployment.
- Returns
- Each answer is a reproducible failure, minimized and saved as a test.
Ordinum is verification infrastructure: a search engine for the failures your system can reach.
- Not a pentest
- Not an observability dashboard
- Not an AI wrapper
- Not a static scenario library