01 / Specify
Record scope, acceptance criteria and permissions in a versioned Work Contract.
IN DEVELOPMENT
AI can do the work.
Testudax helps prove it was done right.
Testudax is developing an independent assurance platform for AI-delivered work — starting with software. Define acceptance criteria, verify submissions using reproducible checks and review evidence before acceptance.
A clearer definition of done
AI-generated output is not the same as accepted delivery. Testudax makes requirements, verification and remediation explicit before work begins — a versioned Work Contract, an independent verifier and an evidence record your team can inspect.
02 / Submit
Use work delivered by a compatible AI coding agent without replacing your preferred tool.
03 / Verify
Evaluate agreed requirements using reproducible checks and independent evidence.
04 / Decide
Review outcomes, accept valid work or require remediation.
What exists today
The first verification slice runs on a live deployment. Recorded walkthroughs show a contract commissioned, a submission verified — including a deliberately defective one being rejected — and a signed evidence bundle accepted, all through the deployed UI.
DELIVERED Independent verification
Bound probes run acceptance criteria in isolated containers against the submitted artefact. Producer and evaluator are separate identities.
DELIVERED Signed evidence
Verdicts are issued as signed evidence bundles (DSSE envelopes) with a hash-chained audit export — inspectable offline, not a badge.
PLANNED Network & settlement
Interoperable agent network, managed execution and optional settlement rails remain in development or design.
Help shape the product
If your team uses AI coding agents, we are interested in understanding how you define, review and accept their outputs.