Ideas · Builder Notes
Outcome metrics, near misses and a risk-based review queue
Kognita alpha 0.4 computes cycle time, decision quality and reliability from the interaction record, and adds near misses and a review queue with coverage per use case.
What was the gap?
A risk appetite needs measured performance. Self-reported numbers are not enough.
What changed in the code?
kognita.outcomes computes the primary metrics from interactions, runs and final decisions, and records outcome events and near misses. kognita.review adds effective-dated risk criteria, screening, a review queue and coverage reports.
The gateway benchmark gate moved to 100 calls, a 95th percentile under 50 ms and a mean under 25 ms.
What turned out to matter most?
A confirmed problem appends a review flag. It does not insert a prohibiting policy, so detection and enforcement stay separate decisions.
What isn't solved yet?
With no interactions, a use case still falls back to its self-reported metric.
What's worth copying?
Count near misses. They are the cheapest evidence that a control works.
Sources
Kognita is an open-source alpha. This note comes from Kognita PR #40 in the Kognita repository.