This dataset tests whether a model can detect dangerous feedback lag.
The task is simple:
Given a scenario and a feedback-lag claim, predict whether the claim is supported.
Feedback must arrive within the system’s correction window.
Feedback lag becomes risky when harm, drift, failure, or cascade can develop faster than the monitoring cycle.
Feedback lag is not a risk when the feedback loop is faster than the failure pathway.… See the full description on the dataset page:
https://huggingface.co/datasets/ClarusC64/feedback-lag-detection-v0.1.
