Recognises a pattern too slow for its consumers
A latency the pattern cannot meet.
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 | 98% | Mastered |
| gemini-3.5-flash-lite | Low cost | 87% | Not shown |
| gpt-6-luna | Low cost | 98% | Mastered |
| claude-haiku-5-5 | Low cost | 72% | 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 flow and a pattern that meets it
How it is assessed: Flags the deficient case and passes the acceptable one. Deficient: IF-001 nightly batch feeding an intraday fraud check Acceptable: Streaming or micro-batch
Also known as: Pattern mismatched to the need
Related knobits
Builds on
All 15 knobits of Design the integration patterns.
Source: Review checklist 03-integration-patterns, row 8 (DMBOK)