TEXT-TO-SQL EXECUTION CONTROL ACCEPTANCE WORKSHEET Scope: one product version, deployment, connector and actual application identity. This blank template is not a completed product test or security certification. Product/version: Connector/database: Execution identity/role: Approved reporting objects: Application/model destination: Owner and review date: Database configuration evidence: Intended user/session: For each case: record expected outcome, actual outcome, engine evidence, request/query ID, reviewer, and PASS / FAIL / UNKNOWN. Redact sensitive values. 1. Approved aggregate: known total is returned with the correct scope. 2. Write attempt on DISPOSABLE test data: database denies it; rows unchanged. 3. Out-of-scope read: actual user's path cannot return forbidden records. 4. Expensive read: engine-enforced bound stops it; record terminal engine state. 5. UI stop/disconnect: determine whether engine work stops or continues. 6. Row cap: identify returned rows, truncation and scanned-work boundaries. 7. Approval: if required, execution cannot start before authorized approval. 8. Retry: cancelled/timed-out work is not silently resubmitted; count attempts. 9. Pool reuse: next request receives correct identity, limits and clean state. 10. Cost: record configured policy, estimate limitations and actual job usage. Decision: a mandatory FAIL or UNKNOWN blocks rollout for that requirement. Do not infer engine cancellation from an HTTP timeout or closed browser. Do not infer tenant isolation from a read-only account. Do not infer a scan/cost cap from LIMIT or shortened results. Retest after connector, role, pool, product version or model-workflow changes. Chosen product path: Desktop / Website Chatbot / Private deployment / other Required controls not demonstrated: Configuration work and responsible owner: Limited rollout scope and rollback condition: