Calibrated Identity

Calibrated Identity is the patented entity resolution capability inside Concord by IaxaI. Most ER systems output a similarity score that means whatever the model felt that day. Concord outputs three numbers stacked together. The Bhattacharyya coefficient measures raw distribution overlap between the two values being compared, per entity category. Temperature scaling turns that coefficient into an honest per-category probability. Conformal prediction adds a coverage-guaranteed set: `{match}`, `{non-match}`, or `{match, non-match}` when the math says we genuinely don't know. The decision tier (auto-approve, human-review, auto-reject) keys off the calibrated probability with thresholds the operator controls. Downstream surfaces (Semantic Alert Dedup, the knowledge graph, Detection Portability) read the calibrated number as a prior so a shaky identity claim never anchors a high-confidence narrative. The patent is in active prosecution at USPTO. The math is auditable end-to-end and every match writes its full rationale to the ledger.