You don't win the model race.
You sit under it.
Everyone is spending billions to make their proposer smarter. The floor is the layer they all still need — the one that makes any proposer deployable, auditable, and insurable. We're not betting on the winner of the intelligence race. We're the runtime the winner still has to run on.
Perslis is not the driver.
It is the brakes.
Perslis is the safety and control harness between probabilistic intelligence and actuation. A planner — any model, yours or ours — may propose “change lanes” or “accelerate”. A deterministic layer then decides whether that action is admissible given the current state, the constraints, the permissions, and the safety invariants. If it is not, it never reaches the actuator.
The model cannot bypass the floor.
Every command passes through the admission layer before actuation; there is no side channel to the actuator. A hostile or broken planner is exactly the test.
Unsafe commands are rejected deterministically.
Same state, same command, same verdict, every time — with a receipt naming the invariant that fired. Not a probability, not a policy: a rule you can read.
Insufficient confidence or state → a known safe state.
When the world state is stale, the sensors disagree, or the planner's confidence drops below the bar, the floor does not guess: it drives to the defined fallback — slow, stop, hold.
Models can drive. We're building the brakes.
Read the full explanation — every step from proposal to actuation →
The upsell is simple: bring your model, add the brakes. Capability stays yours; admissibility, refusal and the safe fallback become a verified layer under it, with an audit trail.
Why a floor is a company,
not a feature.
1 · Model-agnostic by construction
We're not a competitor to the autonomy stack — we're the runtime beneath it. The safety case never rests on whose model is best, so we sell to every team building an autonomous system, not against them.
2 · The moat is the runtime
The defensible asset is the admission runtime and the receipt trail it produces — the auditable, certifiable safety argument. Models commoditize on a brutal curve; a verified enforcement layer with an inspectable record does not.
3 · Provider- and hardware-agnostic
One abstraction, many floors — car, drone, robot, boat, bike. Any proposer, any actuator. The same engine and the same receipt shape port across verticals, so each new domain compounds the platform instead of forking it.
4 · De-risked by design
We solved the hard software-control question first, in simulation: given a trustworthy read of the world, what motion is admissible — enforced independent of the model. The physical layers are the fundable roadmap, not an unproven leap.
One admission runtime,
three domains.
Motion isn't one vertical. It's the third instance of a single primitive Perslis already runs in science and law: an admission-control runtime for model-mediated systems. The same question — can this enter trusted state? — asked of a claim, a citation, and an action.
Science
The bio floor rejects fabricated protein facts; only source-pinned claims are admitted.
Law
The legal floor rejects fabricated case law; a citation must resolve to a real record.
Motion
The motion floor rejects unsafe commands; only admissible actions reach the hardware.
A horizontal platform, not one vertical: wherever a model proposes and the world pays for mistakes, the floor is the layer that decides what's allowed through.
Models commoditize.
The runtime doesn't.
Models commoditize. A verified admission runtime with an auditable receipt trail does not.
What exists today,
stated plainly.
What exists today is a software runtime-safety architecture demonstrated in simulation (MetaDrive) — not a certified real-vehicle system. The physical layers — perception, latency, actuator faults, and certification — are the roadmap, and we say so up front.
That's deliberate. Those layers are downstream of the question we've already answered: given a trustworthy read of the world, what motion is admissible — enforced independent of the model? We solved the control question first because it's the part that doesn't get easier with a bigger model or better sensors. The rest is engineering with a clear line of sight.