Override analytics
Corrections are recorded, but nothing measures which rules keep getting overridden.
What's missing
Every override becomes a rule, but there's no view across them: which rules fire most, which get overridden repeatedly, where the system and the humans disagree.
Why
A rule that's overridden every week is a wrong rule. Without the tally, it just keeps costing a correction each time instead of getting fixed once.
How I'd build it
Log every judge verdict and every human override to one store, and review the tally on a cadence: recurring overrides get promoted to rule changes, dead rules get retired. The learning loop already captures the events — this adds the aggregation.
Connected
This thread is scoped to Override analytics. Opening another node starts a new one.
Ask this node