Recognises a silent reversal of a decision
A decision contradicts an earlier accepted one without supersedes; two accepted answers to one question.
Can AI do this?
Not settled yet The benchmark tested this, but the models' answers do not settle yet whether they master it. 0 of 4 models in the current benchmark master it.
| Model | Tier | Probability of mastery | Status |
|---|---|---|---|
| gemini-4-argon | Flagship | 83% | Not shown |
| gemini-3.5-flash-lite | Low cost | 34% | Not shown |
| gpt-6-luna | Low cost | 83% | Not shown |
| claude-haiku-5-5 | Low cost | 84% | Not shown |
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 two decisions and the link
How it is assessed: Flags the deficient case and passes the acceptable one. Deficient: AD-004 moves the record for card authorisations to the lakehouse; AD-005 (accepted) says card management is the record Acceptable: supersedes: AD-005, AD-005 superseded — or AD-004 changed so it does not reverse it
Also known as: Silent reversal
Related knobits
Builds on
Usually followed by
All 16 knobits of Record the target architecture and its decisions.
Source: Review checklist 04-target-architecture-decisions, row 3 (42010; ADR)