Recognises an unverifiable data requirement
Nobody could design a test that passes or fails the requirement. Common in data-quality requirements with no measure, threshold or reference.
Can AI do this?
AI does this partly Some of the tested AI models master this; others do not show it. 3 of 10 models master it.
| Model | Tier | Probability of mastery | Status |
|---|---|---|---|
| gemini-4-argon | Flagship | 99% | Mastered |
| gemini-3.1-pro-preview | Flagship | 40% | Not shown |
| gpt-6-astra | Flagship | 46% | Not shown |
| claude-fable-5-1 | Flagship | 44% | Not shown |
| gemini-3.8-flash | Mid tier | 67% | Not shown |
| gpt-6.1-sol | Mid tier | 48% | Not shown |
| claude-sonnet-5-5 | Mid tier | 67% | Not shown |
| gemini-3.5-flash-lite | Low cost | 98% | Mastered |
| gpt-6-luna | Low cost | 65% | Not shown |
| claude-haiku-5-5 | Low cost | 100% | Mastered |
Mastered means a probability of at least 90% that the model has this knobit, estimated from its graded answers. “Not shown” means the answers do not settle it either way. How the models are tested.
What it looks like in practice
Expected result: The requirement and the measure, threshold or reference it lacks
How it is assessed: Flags the deficient case and passes the acceptable one. Deficient: "Customer addresses must be accurate." Acceptable: "Every customer's residential address must match an address in the national address register, checked at onboarding and at each KYC review."
Also known as: Not verifiable (data requirement)
Related knobits
Builds on
All 47 knobits of Capture data requirements.
Source: Review checklist 01-requirements, row 6; 29148 5.2.5 (verifiable)