# AWS describes Adjudicated Query for lease compliance reviews

- Source: MechNext — https://mechnext.eu/en/article/30-aws-describes-adjudicated-query-for-lease-compliance-reviews
- Published: 2026-10-05
- Updated: 2026-10-05
- Category: Industry News
- Language: en (Markdown in other languages: ?lang=de, ?lang=sk, ?lang=pl, ?lang=zh, ?lang=ru)

> An AWS reference pattern puts a conversational Amazon Quick interface over deterministic rules, while keeping population selection and compliance determinations outside the language model.

An October 2, 2026 post on the AWS Machine Learning Blog describes the Adjudicated Query pattern for asking compliance questions through an Amazon Quick chat interface. Lease compliance is the running example. The design separates the natural language people use to ask questions from the mechanism that evaluates predefined rules.

The conversational layer has a bounded role in the described pattern. The model translates a question into a call to a fixed, typed operation, then narrates the returned result. It does not write queries, select the population being checked, or decide whether a case is compliant. That division matters: a natural-language answer is an interface to results produced by the rules tool; it is not itself the compliance determination.

### Determinations stay with the rules engine

Adjudicated Query pairs a chat interface with a deterministic rules engine. A user can phrase a question conversationally, while the system maps it to one of the available operations with defined types. The rules engine returns results for the conversational layer to explain. This lets people ask in ordinary language without having the model invent queries or judge lease clauses on its own.

The post also describes a completeness receipt. Before results persist, it asserts that the counts for compliant, in-breach, ambiguous, and unreadable records add up to the scanned total. The receipt therefore concerns how the processed population is distributed among those stated categories. By itself, it does not establish that the rules are correct, that input data are accurate, or that the population supplied for review covers every relevant record.

### Reference architecture

In the reference architecture, MCP requests pass through Amazon API Gateway and AWS Lambda to Amazon Aurora Serverless v2. Amazon Quick Sight reads the same Aurora store through a VPC connection. The post thus describes a request path for the operations and a shared data store that also feeds the results dashboard.

Amazon Bedrock has a narrower role in this account: it is used for exploratory clause search and is not consulted for official compliance determinations. This keeps rule evaluation separate from search that can help locate relevant text. The post does not describe Bedrock as the adjudicator, nor does it present chat output as an independent legal assessment.

### What the sample reports

In the illustrative example, a Texas late-fee-cap sweep reports 10,111 violations, 689 ambiguous records, and 20 unreadable records. It is presented alongside a completeness receipt and a synthetic-data caveat. These are figures from the example, not findings about actual customer leases. The post says the sample ships with synthetic lease data and placeholder rules and citations, so the example should not be read as validated legal guidance.

The post also describes what users see in the dashboard. Each row represents a lease-rule pair and includes the lease ID, the rule that fired, the extracted value, and the expected value. A finding detail also shows the lease clause and information about the rule. These fields connect an outcome to a particular lease and check, giving readers context for what the displayed result refers to.

### Potential areas of use

AWS uses lease compliance as its main example and names insurance claims adjudication, sanctions screening, and export control as other domains where the pattern may apply. In such settings, the described approach can expose predefined checks through questions phrased in natural language. Its operation still depends on the defined operations, rules, and population provided to it.

The design’s central idea is a division of labor: conversation helps a person ask and understand, while a deterministic mechanism evaluates fixed rules. The receipt expresses a count check across categories in the scanned population, and the displayed rows link leases with rules and values. Results should therefore be understood within the scope and assumptions of the particular review. The architecture description and illustrative figures do not guarantee real-world correctness, legal compliance, or coverage beyond the population that was scanned.