Dubitor
When an AI-backed decision is about to become policy, it has outside people examine it. The judgment is signed and shows where the reasons give way.
EnterResearch
One method, one trajectory, several instruments. The four steps to put a number or a decision to the test; the path that leads from verification to precaution; the instruments that apply the method to different objects.
The path along which cognitive sovereignty extends: from the number that can be checked against the real to decisions where error carries consequences we cannot afford.
Here an estimate can be checked against the real. Numbers are verified, uncertainty is made explicit, and it becomes clearer when a forecast earns trust.
Ordinary frequency is no longer enough. A system is put under stress to find where its assumptions give way.
The future has not happened yet. Hundreds of possible trajectories are explored, to see which consequences emerge and which scenarios do not hold.
When consequences are irreversible, the probability of harm leaves the decisive question open: which exposure are we willing to accept. Here verification becomes infrastructure for the precautionary principle.
Four steps, always the same, whatever the object: a number, a decision, a plan. The method comes before the instruments: they apply it to different objects.
01
Where does it come from, what has to be true?
Before judging an output, you put it in writing: data, version, assumptions.
02
How often has it been right so far, really?
The distance between promise and outcome, measured on what has already happened.
03
Where does it stop holding?
The point where the system gives way and the error becomes irreversible. The boundary matters more than the probability.
04
Who answers if it is wrong?
A name on top of the decision. After the machine, the responsibility goes back to a person.
They come from the same thesis: verifying what a machine produces before it becomes a decision. Dubitor works on the reasons behind a decision; Vektor, Roundel, and Probe on estimates, breaking points, and action scenarios.
The method stays the same; the object of verification changes. This is the direct mapping.
When an AI-backed decision is about to become policy, it has outside people examine it. The judgment is signed and shows where the reasons give way.
EnterOne private beta and two research lines: risk estimates, structural fragility, scenarios before action.