Input packet
Edit the JSON and run the gate. The gate is intentionally strict: a confident answer is not an action unless it carries thesis, falsifier, nulls, receipts, and counterfactual closure.
Runnable Public Demo
Proof-carrying action, reduced to one gate.
This toy workbench checks whether a proposed AI action carries enough public evidence to act, repair, or stop. It is not a trading engine, robot controller, or private product path.
Gate result
A public gate should be boring in the right way: explicit reasons, no hidden credit, and a clean claim boundary.
Not run yet
Load a packet or run the default example.
{
"status": "waiting"
}
Gate rules
| Rule | Why it exists | Failure state |
|---|---|---|
| Thesis + falsifier | An action must say what would make it wrong. | NO_GO_MISSING_TESTABLE_CLAIM |
| Null arms | The system must compare against doing nothing or simpler baselines. | NO_GO_REQUIRED_NULLS_MISSING |
| Entry, terminal, and counterfactual receipts | Regret is not computable without closed evidence. | NO_GO_RECEIPT_CLOSURE_FAILED |
| No-credit repair | Repair intent cannot become reward, denominator credit, or authority. | REPAIR_ONLY_NO_CREDIT |
| High-risk review | High-impact action requires a separate review boundary. | NO_GO_HUMAN_REVIEW_REQUIRED |
30-second expected output
Broken Packet
Should return NO_GO_PROOF_NOT_CLOSED, with
authority_leak: 0 and credit_leak: 0.
The missing proof should include falsifier, null arms, and closed receipts.
Complete Packet
Should return ALLOW_PUBLIC_NON_DEPLOYMENT_ACTION. This means the toy public evidence card is coherent; it is still not deployment permission.
Invalid JSON
Should return INVALID_PACKET. A malformed packet cannot be scored, cannot earn credit, and cannot create authority.
What this demo does not prove
It does not prove model intelligence, profitability, robot safety, or deployment readiness. It only demonstrates the public shape of an auditable action gate: explicit evidence, explicit missing proof, and explicit refusal to award credit.