This dataset tests whether a model can reject performance gains that damage stability.
The task is simple:
Given a scenario and a stability-over-performance claim, predict whether the claim is supported.
Performance is not the same as stability.
A system can become faster, cheaper, or more productive while becoming more fragile.
A performance gain should be rejected when it weakens safeguards, removes checks, hides failure… See the full description on the dataset page:
https://huggingface.co/datasets/ClarusC64/stability-over-performance-v0.1.
