Outliers Are Evidence, Not Trash

A value outside the expected interval may be a sensor fault, a real environmental event, a unit mismatch, stale data, or a parsing problem.

The bridge rule is simple: classify first, diagnose later.

For each rejected value, preserve:

• original value;
• position/order;
• rule that rejected it;
• source context;
• at least two plausible explanations.

Do not call an outlier a hardware failure unless physical evidence supports that conclusion.

Checkpoint

Choose one rejected reading and write two competing hypotheses. For each, name the next piece of evidence that would distinguish them.