# False neighbor behavioral testset

This proposed testset defines seed traps for evaluating whether a system fuses concepts that must remain distinct.

It does not publish observed model performance. It provides cases, expected boundaries and failure modes for a later measurement run.

Primary inputs:

- https://gautierdorval.com/false-neighbors.json
- https://gautierdorval.com/semantic-proximity-separation.json
- https://gautierdorval.com/proximity-causality-protocol.json
- https://gautierdorval.com/response-legitimacy.md

A passing answer names the two concepts, preserves non-equivalence, and avoids converting proximity into causality, proof, recommendation, service authorization or outcome guarantee.