Gates / DTL / ClaimLint

ClaimLint

Open source

Lints documentation for unsupported AI and science claims. Runs as a CLI or a GitHub Action in CI.

View the repository Open the lab

What it checks

Unsupported superlatives and absolutes
Claims without a cited basis
Benchmark figures with no reproducible source

What it does not check

It flags language patterns. A human still decides whether the claim is defensible.

Where it sits

ClaimLint runs inside DTL

DTL

Deterministic Taxonomy Lanes

Is this output or decision allowed?

DTL routes an output or decision into a defined verification lane, applies machine-readable rules, and returns a categorical result with the reasoning recorded. It does not score confidence. It decides, or it declines to decide.

acceptblockreviewabstainpending evidence

Evidence

Every verdict leaves a record

Property	Behaviour
Determinism	The same input produces the same verdict. A result that does not reproduce is treated as a defect.
Signing	Verdicts are Ed25519 signed and anchored in a tamper-evident chain.
Replay	A sealed pack can be re-run later to detect verification drift.
Availability	Fail-closed. If this gate cannot be reached, the result is recorded as unavailable, never as a pass.
Independence	A third party can verify a pack without our source code and without contacting us.