Did the database write match its approval?

Compare the request, execution record and audit event

An agent gets approval to update a database. The application reports success. Later, someone reads a different value. Did the approved write happen, or did another write change it afterwards?

Try the reader and examples. They compare the approved SQL and parameters with the application record and audit projection, including the database, schema, principal and transaction outcome. A later state change stays separate from an earlier matching commit.

Cho HyunJong brought this case to the Lab thread. The examples are ours. A real Tadpole approval and database audit pair is the next contribution; another project can bring a reader and compare results.

Want to help? Bring a shareable pair, run the examples with another reader, or pick a check to keep in your own CI. Join the case in the Lab. We maintain the comparison code and credit the case, code and runs separately.

Run it and inspect the exact fields, source and results

The illustration has 16 author-created fixtures and 56 passing tests. It compares supplied records; it does not authenticate their source or verify that an audit collection is complete.

From the Atlas checkout:

cd examples/approved-sql
uv sync --frozen --group dev
.venv/bin/python -m sql_illustration.check fixtures/matching-commit-later-change.json
.venv/bin/python scripts/qualify.py

The first command reports recorded_commit_matches alongside supplied_state_differs. Changed SQL, parameters, target or identity produce separate failures. Missing or ambiguous audit evidence keeps its own outcome.

Read the input contract, source map, design decisions and retained check record. Claims P-83 and P-84 bind the source and measured scope.