Legal says one unsupported claim in a brief is an incid...

The conflict arises because the team is arguing over the wrong axis. The requirement for Pr(S=1) can be met by reducing the number of claims (n) in each brief, which makes the threshold easier to hit. Alternatively, gating each claim individually turns the problem into a binary per-claim judgment, which is easier to manage. This approach may cost recall, so you should commit to maintaining F1@K at a fixed K to ensure the trade-off is transparent.