Writing · Topic
Evals and testing
Evals measure whether an AI system is right often enough to trust, offline and online. We write about what to actually evaluate, and why it stays an ongoing retainer rather than a one-time gate.
Writing topics8 in the taxonomy
Is 95% accuracy good enough, and who signs it off?
No percentage is good enough on its own, and the sign-off belongs to the owner who carries the consequence of a wrong value, not the engineer and not the vendor.