Known repeats caught: 24 / 30 (80% recall). All 60 reviewed items: 24 repeats caught, 6 missed, 30 correctly new, 0 false alarms. 12 synthetic calls over 4 weeks.
What this measures: repeat matching on 12 synthetic calls. Preliminary evidence, not an untouched holdout, a medical result, or a real-call performance estimate.