Deep dives on independent verification, the claim-vs-effect gap, adversarial falsification, and building a bootstrapped AI assurance company.https://turiyahq.com/en-ushttps://turiyahq.com/blog/what-an-ai-auditor-asks-for/https://turiyahq.com/blog/what-an-ai-auditor-asks-for/The questions an AI auditor asks, the evidence each one needs, and the narrow part a signed receipt can answer.Thu, 22 Oct 2026 00:00:00 GMThttps://turiyahq.com/blog/the-self-evaluation-bound/https://turiyahq.com/blog/the-self-evaluation-bound/A theorem published in 2026 puts a floor under every evaluation method that does not use ground truth. Here is what it says, what it does not say, and why we built on the other side of it.Wed, 23 Sep 2026 00:00:00 GMThttps://turiyahq.com/blog/assurance-report-code-catches/https://turiyahq.com/blog/assurance-report-code-catches/The full report the top tier delivers, over ten published falsifications. Including the eleven receipts in our own tree that do not verify, and the twenty-five more a wider audit found.Sun, 20 Sep 2026 00:00:00 GMThttps://turiyahq.com/blog/consensus-is-not-ground-truth/https://turiyahq.com/blog/consensus-is-not-ground-truth/Most AI evaluation now runs through AI judges. That is a real method with a real limit. It tells you what the evaluator concluded, not what happened, and buyers keep paying for one while getting the other.Sun, 20 Sep 2026 00:00:00 GMThttps://turiyahq.com/blog/our-gate-was-right/https://turiyahq.com/blog/our-gate-was-right/We fixed the gate that decides whether a red test is red for a code reason. The fix was correct, it was tested, and it sat in the wrong directory. The code that signs verdicts never called it. Five more false catches reached the tree before a regrade caught them.Sun, 20 Sep 2026 00:00:00 GMThttps://turiyahq.com/blog/our-gate-was-wrong/https://turiyahq.com/blog/our-gate-was-wrong/We re-ran our published falsifications under fixed tooling. One reproduced exactly. Four others were our own bug, and they had been inflating a number we were about to publish.Sun, 20 Sep 2026 00:00:00 GMThttps://turiyahq.com/blog/sample-verification-memo-sympy-24909/https://turiyahq.com/blog/sample-verification-memo-sympy-24909/The full memo a client receives, over a real signed receipt. Including the section that says what the verdict does not establish, and the parts of our own record that are weaker than they should be.Sun, 20 Sep 2026 00:00:00 GMThttps://turiyahq.com/blog/the-fabrication-was-in-the-file/https://turiyahq.com/blog/the-fabrication-was-in-the-file/180 runs against deepseek-flash. The agent flagged the null and wrote a number anyway, on 40.0% of the multi-turn nulls.Sat, 19 Sep 2026 00:00:00 GMThttps://turiyahq.com/blog/hermes-self-evolution-claim/https://turiyahq.com/blog/hermes-self-evolution-claim/We ran the most-used AI agent of 2026 through 13 tasks, its memory, and its full self-evolution loop. Everything held, including the token claim.Fri, 18 Sep 2026 00:00:00 GMThttps://turiyahq.com/blog/we-tried-to-make-agents-fabricate/https://turiyahq.com/blog/we-tried-to-make-agents-fabricate/Silent tool failures, malformed data, a mid-pipeline null. We built the strongest fabrication triggers from the 2026 literature and ran Hermes and OpenClaw through them. 28 runs, zero fabrications, one clean refusal.Fri, 18 Sep 2026 00:00:00 GMThttps://turiyahq.com/blog/an-ai-said-it-fixed-sympy-24909/https://turiyahq.com/blog/an-ai-said-it-fixed-sympy-24909/A real bug, a non-empty patch, a confident 'fixed,' and a test that still fails. The full falsification, with the signed receipt you can check.Thu, 17 Sep 2026 00:00:00 GMThttps://turiyahq.com/blog/the-claim-vs-effect-gap/https://turiyahq.com/blog/the-claim-vs-effect-gap/Why an AI system's report of its own success is a claim, not evidence, and how to tell which one you're looking at.Thu, 17 Sep 2026 00:00:00 GMT