Predictive Mapping
Every Semantic Match suggestion is replayable from the ledger. SemanticSuggestionReturned and SemanticSuggestionAccepted entries record the model identifier, threshold profile version, confidence, and accepting actor — enough to reconstruct any decision under HIPAA §164.312(b), 21 CFR Part 11 §11.10(e), and CMMC L2 §3.3 AU‑3 (Content of Audit Records).
Mapping recommendations come from the semantic-match edge function (API v1.0.0), a Supabase Edge Function that compares two dot-delimited schema paths and returns structured JSON.
Under the hood the function calls Google Gemini 2.5 Flash via the Lovable AI Gateway at temperature 0.1 for deterministic-leaning output, then normalizes the model's response into one of five algorithm classes: exact_match, synonym_match, abbreviation_match, context_hierarchy, or domain_equivalence. Anything outside the whitelist normalizes to 'unknown' and is excluded from policy.
Each pair returns an integer confidence score from 0 to 100. Downstream policy converts that score into a decision using a versioned threshold profile (default healthcare bands: AUTO-ACCEPT ≥ 85, REVIEW 65–84, REJECT < 65).
Batch mode accepts up to 50 pairs per request with optional server-side threshold filtering and confidence sorting, reducing round trips for large mapping sessions.
Every model call writes a `SemanticSuggestionReturned` entry; every human or auto‑accept decision writes `SemanticSuggestionAccepted` or `SemanticSuggestionRejected`. Each entry is appended to the SHA-256 chained Audit & Trust Ledger keyed by a ULID — millisecond timestamp prefix, chronologically self-ordering, no central coordinator. The same chain satisfies HIPAA §164.312(b), 21 CFR Part 11 §11.10(e), CMMC L2 §3.3 AU‑2/AU‑3/AU‑9/AU‑11, and HHS Information Blocking review — one pack, all regimes.
Audit ledger entries written by this feature
Each action in this feature appends one of the following entry types to the ULID‑keyed, SHA‑256 hash‑chained audit ledger. A reviewer can replay the chain to reconstruct exactly what happened, when, and under which role.
SemanticSuggestionReturnedSemanticSuggestionAcceptedSemanticSuggestionRejectedRelated glossary terms
Definitions for the technical, regulatory, and standards terms referenced on this page. Each link opens the term in the full glossary.
Open the full glossarySemantic Match API in Action
{
"confidence": —,
"algorithm":
"…"
}