Lost claims represent substantive facts and propositions from a source webpage that were relevant to the target query but were completely omitted by the model in its grounded answer, reducing the overall aggregate fidelity score.