Long tail data

Examples of less frequent conditions or events within the distribution a system needs to handle.

Rare does not automatically mean useful. Prioritize cases that matter to the intended task, such as unusual interactions, recoveries or ambiguous outcomes. Record how the examples were selected: a deliberately difficult collection should not be treated as a representative estimate of everyday performance.

In practice

Mining identifies uncommon clips where a part initially seats and then backs out after release.