Database Architect Analysis
The architect tools are designed for review, not automatic schema changes. They make the distinction between SQL Server facts and semantic interpretation visible in the result.
Deterministic review
review_database_schema reads a catalog snapshot. It reports missing primary
keys, disabled or untrusted foreign keys, foreign keys without a leading
unfiltered index, and conservative possible missing relationships. It also
derives declared relationship cardinality and optionality from keys and
nullability. Findings carry catalog evidence so you can verify them in SQL
Server. A finding is a review prompt, not proof that a constraint or index
should be added.
Bounded data quality signals
profile_table_data accepts explicit table and column selections. It samples
within configured limits and returns null counts and, for supported types,
distinct counts. It does not return source rows. The sample has no ordering
guarantee, so its counts are not whole-table quality claims.
Entity-pair analysis with Jev
analyze_entity_pair compares two selected tables and always keeps local
findings and declared relationships available. Semantic analysis is a separate
per-call opt-in and requires Jev to be enabled in server configuration. Its
outbound state contains selected names, types, nullability, keys, and
relationships. Optional profiling can add aggregate counts. It does not send
source row values, SQL definitions, connection details, or credentials to Jev.
Jev answers can suggest whether two tables express the same concept, whether a relationship seems missing, or what entity pattern they resemble. The answers include a model version and are advisory. If Jev is unavailable, the local findings remain in the result. The tool does not generate or apply DDL.
Current evaluation status
An initial evaluation used eight provisionally labeled table pairs from AdventureWorks and WideWorldImporters. The semantic scores overlapped between some positive and negative missing-relationship cases, and one negative pair was repeatedly classified as a duplicate concept. No automatic review threshold has been selected. Read the evaluation report for the case set and limitations.
The longer-term goal is a database architect that can discover entities, review normalization, detect duplicate concepts, analyze data quality, and recommend model improvements. Those workflows need more labeled examples and evidence rules before they can be treated as reliable outputs.