AI Demos Research — the structured-intelligence platform. Every verdict on these pages opens to the execution behind it.
AI Database Agents v1AI for Databasefew-shot-generalizationanswer
AI for Database

The user supplies example questions and the queries they trust

untested

Verified queries, question/SQL pairs, few-shot examples. Whether they are used — and whether they GENERALISE. A product that answers the exact supplied question but fails a rephrasing has been given a lookup table, not trained.

reading this cell
As ai-database-agents v1 reads it — only grades under the rubric that version pins.
reading policy v1 · fingerprint 4f53cda18c2baa0c…

What we measured

Nothing was measured numerically for this assessment.

Every assessment

0 counting · 0 from captured bytes · 0 transcribed

RefVerdictRubricProvenanceGraded byCounts here 

Nobody has assessed this tool on this question. `untested` is what the page says, and it is a value rather than a blank.

Every verdict on these pages is openable to the execution behind it. Where an execution was reconstructed rather than captured, the page says so — a finding and a measurement are not the same claim.