Only transactions selected for review receive confirmed fraud labels. The team wants to label all unreviewed transactions clean. Explain the problem and a first audit.

The labels are selectively observed under the review policy. Unreviewed is not confirmed negative. Audit selection rates, policy changes, label maturity, and feature distributions. Define a prediction target supported by the data or a justified collection strategy. Do not promise that reweighting solves the issue without the required probabilities and assumptions.