Spurious counter examples

[shreyas-jadhav]

Hi! First of all, thank you for creating Polygon, we've been using it for text-to-SQL validation.

We noticed something that we're not sure how to interpret, and we wanted to ask whether this is expected behavior.

In several cases, Polygon returns NEQ along with a counterexample database. However, when we execute both SQL queries on the provided counterexample, the results appear to be identical, so the queries don't seem to be refuted.

Is there something we're missing about how the counterexamples should be interpreted? For example:

• Is the counterexample intended to be used in a specific way?
• Are there any assumptions or execution semantics that differ from standard SQL execution?
• Could this be due to unsupported SQL features or a known limitation of the tool?

If this is unexpected behavior, we'd be happy to provide a minimal reproducible example.

Thanks for your time, and thanks again for your work on Polygon!