01 / INPUT
Send the claim or structured surface.
Try a machine-generated mathematical or factual claim in the live checker, or give the engine a sequence whose rule needs to be recovered.
Proof gates for AI outputs
Make AI outputs earn the right to be trusted.
Chiron checks the claims it can prove exactly, refutes what is false, and refuses the rest. Every result carries the evidence and the boundary of what was not certified.
Use the public demo for non-sensitive test inputs only. It is a live service, not a general-purpose truth oracle.
22 / 22published stamped outputs correct in the current external eval
0false stamps in that published frozen eval
12 refusalsbecause abstention is preferable to an unsupported stamp
What the product does
A decision gate—not another confidence score.
Use Chiron when a system needs to distinguish a claim it can support from one it cannot. It is built for exact, inspectable checks and explicit refusal—not plausible-looking output.
02 / VERDICT
Get VERIFIED, REFUTED, or REFUSED.
Chiron checks only the portion it can handle exactly. It does not bless the surrounding prose just because one statement passed.
03 / EVIDENCE
Keep the reason, not just the answer.
Results include the discovered rule or claim-level status, a machine-readable evidence record, and the limits of coverage. Sequence results also render an Evidence Trace: what the recovery saw, what was held out, and what earned a certificate. Try it live →
The important constraint
Honest scope is a feature, not a footnote.
A system that says “verified” when it only recognized part of the input is worse than no system at all. Chiron surfaces coverage and leaves uncheckable material unblessed.
It can checkSupported exact claim types and structured patterns, with an explanation of the verdict.
It will refuseInputs outside its proof methods, ambiguous evidence, or work beyond the configured budget.
It does not claimTo certify all free-form language, business judgment, legal compliance, or safety in the abstract.
Who gets value now
For teams that need evidence before an output becomes an action.
A strong fit
- AI product engineers who need a gate around supported claims.
- Evaluation owners who need reproducible pass, fail, and abstain records.
- Teams that want a deterministic engine they can run offline under a commercial license.
Not the promise
- A generic hallucination score or a replacement for domain experts.
- A legal, regulatory, or security conclusion about arbitrary text.
- A black-box “trust us” layer that silently turns uncertainty into approval.
Commercial use
Move from a public proof to your own workflow.
Commercial access includes the full, self-contained Chiron engine, the right to run and adapt it for your organization, and a path to contribute changes back without weakening the certified core. Individual, Team, and Business checkout are live.
Can I test it first?
Yes. Use the live claim checker, run the public demo core, and grade the published frozen outputs against public ground truth before buying.
Does the demo see my data?
The browser sequence core runs locally. The claim checker and licensed-engine toggle call a public live endpoint, so use only non-sensitive test inputs.
What happens after checkout?
Checkout collects a requested GitHub username. Delivery is confirmed with the purchaser at the checkout email before a private-repository invitation is granted. See pricing for the exact tier terms.