Ground Truth
The verified correct answer used to train and evaluate a model — the reference reality it is measured against.
Ground truth is the verified-correct label or value for an example — the reference reality a model is trained against and evaluated on. Its quality sets the ceiling on how good (and how measurable) a model can be: you cannot learn or score beyond the accuracy of your ground truth.
Worked example: to train a tumor detector, radiologists label scans as tumor / no-tumor; those labels are the ground truth the model learns from and is tested against. Gotcha: ground truth is often noisier and more subjective than assumed — annotators disagree, labels contain errors, and for many real tasks there is no single correct answer — so a model can be “wrong” against bad labels, and measuring inter-annotator agreement (and cleaning labels) is often higher-leverage than tweaking the model.