<!-- Provide a summary of the bug in the title --> ## Description When computing metrics using the same dataset as ref and pred, the average precision differs from 1. The rest of metrics are 1 as expected. https://aphp.github.io/edsnlp/latest/metrics/span-attribute/