Skip to content
ML Academy

Introduced on day 18

Failure taxonomy

A small, fixed set of categories (data, model, pipeline, world) used to label each observed failure so that recurring root causes become visible.

In detail

Tagging each failure with its category and first test turns one-off firefighting into a list that shows where investment pays: data coverage and definitions, model capacity or objective, training-serving skew and stale tables, or changes in the world such as a new competitor.