Automated compliance checking evaluates submissions — filings, drawings, transactions, records — against machine-readable rules and returns verdicts that cite the governing provision. Done properly it is deterministic: the same facts produce the same verdict every time, AI handles extraction rather than judgment, and anything genuinely ambiguous routes to a human with the evidence attached.
The four-verdict standard
Production-grade compliance automation returns four verdicts, not two: meets, does not meet, information not available, and uncertain — the last being an engineered admission that some provisions require human judgment. A checker that only says pass/fail is guessing at the boundary, and every guess erodes the trust the system exists to create.
Each verdict carries its lineage: the facts extracted, the confidence of extraction, the rule version applied, the provision cited. Amendments are data changes with effective dates, so past determinations stay replayable under the rules that governed them.
Where this is heading
Regulators and standards bodies are actively digitalising codes — building, electoral, financial — into machine-executable formats. Institutions that adopt deterministic checking now inherit each newly digitalised ruleset as configuration, not as a new procurement.