Module 05 Summary

What this module established

Write the edge-case rulings into the schema before any labelling starts - whatever is undecided becomes label noise, and no model sees past it. Measure inter-labeller agreement on a sample: it is a hard ceiling on model performance and takes an afternoon.

Carry forward

• Report the majority-class baseline with every accuracy figure, and ask whether the imbalance is the world's or your collection process's.
• Split by the underlying object. Multiple views on both sides measure memorisation, and the inflation survives cross-validation.
• Check every augmentation against the label it must preserve. A flipped 6 is a 9, and the training loop does not notice.

Before moving on

Move on when your labels have measured agreement, your split is grouped, and every augmentation has been checked against your specific label.

Mark complete