Anonymous explanation comparison

You will compare two explanations of the same recorded classifier result. The evaluation contains 60 cases and may be completed in multiple sessions on this device.

Task: Select the explanation that better helps you understand why the recorded decision was assigned while staying faithful to the evidence. Do not assess whether the predicted operation state is the true state. A deletion probe describes model sensitivity only; it is not an instruction for railway personnel.

Judge factual support and operational advice separately. A longer answer is not necessarily a better answer. Method identities and true labels are hidden.

Work independently. Your anonymous reviewer code is generated automatically and saved in this browser. Use a separate device or browser profile for each reviewer. On a shared browser, start a new reviewer only after exporting the previous reviewer's CSV.