AWS introduces the 'Adjudicated Query' pattern for compliance sweeps

The compliance problem at scale

Imagine a portfolio operator holding 50,000 leases spread across multiple states. Each state updates rules like late-fee caps, notice periods, and security-deposit limits on its own legislative schedule. When a regulation changes, the team has to figure out which leases are now out of line. At small volume a paralegal can do this; at thousands of leases the work moves to software, and a new problem appears: nobody can independently verify the number on the screen.

Two core properties

According to AWS, this problem differs from ordinary enterprise search in two ways:

• Provable completeness: A claim like "we checked all 22,910 Texas leases" must be true and demonstrable. A record never assessed must be reported as unevaluated rather than silently omitted.
• Defensibility: A finding may be challenged months later in litigation, an audit, or a regulatory examination. Defending it means knowing which version of which rule was applied, to which clause text, by what method, on what date, and by whom.

How Adjudicated Query works

The pattern is described as a bounded conversational layer over a deterministic rules engine. The model does exactly two things: translate a natural-language question into a call on a fixed set of typed operations, and narrate the result that comes back. It never writes a query, never fixes the population, and never performs a determination.

Behind the boundary sits a rules engine. Rules are versioned data, not code. The engine knows only generic comparison operators (gte, lte, equals, exists) and contains no branch naming a jurisdiction or topic. A law change is a rulebook row edit, not a code deployment.

The completeness receipt

Every compliance sweep produces a completeness receipt: compliant + in-breach + ambiguous + unreadable must equal scanned. This is computed from counts and asserted before anything persists. A run that can't account for its population never finishes, so there's no path by which a record is silently skipped.

The conversational surface carries only counts, the receipt, and a labeled sample. The full result set, potentially tens of thousands of rows, lives on a dashboard surface reading the same data store and drillable per record. This separation means the model never summarizes away the guarantee.

Why not RAG or text-to-SQL?

According to AWS, semantic retrieval (RAG) offers only a ranked sample with no threshold that means "all of them," making completeness structurally impossible. Text-to-SQL can narrow the population: a hallucinated predicate can silently reduce the scope, and the resulting number looks exact even when the scope is wrong. AWS notes the pattern applies to other high-stakes compliance domains as well, such as sanctions screening, insurance claims adjudication, and export control.