Recognises a review decision contradicting its total
Approved below the threshold.
Can AI do this?
AI does this partly Some of the tested AI models master this; others do not show it. 2 of 4 models in the current benchmark master it.
| Model | Tier | Probability of mastery | Status |
|---|---|---|---|
| gemini-4-argon | Flagship | 90% | Not shown |
| gemini-3.5-flash-lite | Low cost | 89% | Not shown |
| gpt-6-luna | Low cost | 91% | Mastered |
| claude-haiku-5-5 | Low cost | 90% | 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: Rejection, or the findings fixed and re-scored
How it is assessed: Flags the deficient case and passes the acceptable one. Deficient: total: 71, threshold: 80, decision: approved Acceptable: rejected, or the findings fixed and re-scored
Also known as: Decision contradicts the total
Related knobits
Builds on
All 8 knobits of Review the data model against the scorecard.
Source: Review checklist 06-model-review, row 4 (Scorecard; DMBOK §2.3)