Rule health
From the provenance store: for each definition version, how often its cards were dismissed or answered "not clinically supported". A version at or above 30% on either, with at least 5 responses, is listed as needs review. Nothing changes criteria automatically: staff may suspend a version (its cards are no longer shown to clinicians; the coder view shows it as suspended; it still runs and replays) and reinstate it, each with a reason, in the audit trail. A change to the criteria themselves goes through local criteria. Demonstration data.
In this demonstration one staff account can suspend; a site would require the same two approvals as a criteria change. Answered = chose an option, wrote their own, unable to determine, not clinically supported, dismissed, or signed with the card open.
| Definition version | Condition | Offers | Answered | Dismissed | Not clinically supported | State |
|---|---|---|---|---|---|---|
front_hf_ef_type@1.0 | Heart pumping fraction (type) | 12 | 0 | 0 | 0 | fewer than 5 answers |
front_kidney_stage_wording@1.0 | Kidney filtering rate (stage) | 26 | 0 | 0 | 0 | fewer than 5 answers |
front_malnutrition_us_consensus@1.0 | Nutrition (severity) | 2 | 0 | 0 | 0 | fewer than 5 answers |
AI second opinion (monitoring signal, not verification)
Currently off (off by default). When on, each new card that needs a look is sent to the model with the definition in our own words and the evidence lines it shows (no note text); the model answers concurs, does not concur or cannot tell, with a quote from that evidence. The answer is stored in the provenance row and non-concurrences are listed above for human review. It never changes a finding, an option or what the clinician sees.