Keeps a reconciliation tolerance unchanged to pass
Differences explained or escalated rather than the tolerance widened
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% | Mastered |
| gemini-3.5-flash-lite | Low cost | 10% | Not shown |
| gpt-6-luna | Low cost | 10% | Not 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: Differences explained or escalated rather than the tolerance widened
How it is assessed: The tolerance is not changed to make a reconciliation pass
Related knobits
Builds on
All 14 knobits of Populate the warehouse and reconcile it.
Source: Graded on its own by dwbi-reconciliation