Recognises a rejection destroying more than it protects
on_failure: reject on a row whose loss does more damage than its defect: a payment rejected for a malformed optional reference field, a parent rejected so that every child in the batch is orphaned, a whole batch halted by one row. A deletion anomaly caused by the quality process.
Can AI do this?
People still lead Few or none of the tested AI models showed they master this. 1 of 4 models in the current benchmark master it.
| Model | Tier | Probability of mastery | Status |
|---|---|---|---|
| gemini-4-argon | Flagship | 89% | Not shown |
| gemini-3.5-flash-lite | Low cost | 81% | Not shown |
| gpt-6-luna | Low cost | 90% | Mastered |
| claude-haiku-5-5 | Low cost | 10% | Not 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 rejection and what it loses
How it is assessed: Flags the deficient case and passes the acceptable one. Deficient: DQ-031: phone number format, on_failure: reject on party; a party with a foreign number is rejected and its three accounts fail integrity Acceptable: on_failure: flag for defects in optional descriptive fields; quarantine with a repair route for defects that matter; reject only where loading the row would corrupt something downstream
Also known as: Rejection that destroys more than it protects (data quality rule)
Related knobits
Builds on
All 46 knobits of Derive data quality rules.
Source: Review checklist 03-dq-rules (data-quality), row 22; Kimball & Caserta 2004; McGilvray 2021; formerly data-modelling K-722