Harmondale
Definition

Validation data

Validation data is data used to evaluate a trained model during development and tune non-learned choices such as thresholds, prompts, hyperparameters, or stopping decisions. It helps detect overfitting before the final independent test.

Last updated: 25 June 2026

Why it matters

It protects teams from believing a model works simply because it memorized or overfit the training examples.

Signals to watch

  • It is separated from training examples
  • It guides tuning decisions
  • It checks overfitting before launch