The engagement
Turiya is not software you install. It is an independent verification engagement: we take one high-stakes workflow, re-run its claims against real ground truth, and hand you signed evidence you can pass to someone who will not take your word for it.
What we verify
One narrow, checkable thing: a system asserted it did something, and it didn’t. Not “is the model good” (an eval), not “is the system safe” (a red team). Whether the effect matched the claim.
On the system’s own account, a failure and a success are indistinguishable. The patch that did not fix the bug reports that it did. Nothing inside that report separates the two cases, because the report is the thing in question. So we do not read it. We re-run the work.
The unit
| Scope | one workflow, roughly 100 to 500 claims |
| Duration | about two weeks |
| Where it runs | inside your environment |
| What you receive | a signed report, one signed receipt per claim, a summary attestation, and every candidate we rejected |
Who this is for
A financial-services team or an insurer being asked for evidence that an AI system does what it claims, and who cannot supply it, because the only account of what happened comes from the system itself.
Independence is not a preference in that seat. Established audit and testing frameworks place significant weight on independence and impartiality — the independence principles reflected in SOX/PCAOB auditing, ISO/IEC 17025 testing laboratories, and SOC 2 examinations.
Where to go next
- How it runs: what runs where, and what we never hold
- What you get: the deliverables, and who owns them
- Reading a receipt: the verdict lattice, and how to check one yourself
- The record: the catches we have published, and the ones we rejected
Ready to scope an engagement?
START WITH A PILOT