diff --git a/src/proficiency_outcome.py b/src/proficiency_outcome.py index 562deea..6304a64 100644 --- a/src/proficiency_outcome.py +++ b/src/proficiency_outcome.py @@ -72,10 +72,10 @@ def recompute_category( For each row: 1. Re-compute benchmark via ``blend()`` on stored leaderboard / self_eval components. If neither component exists the row skips the conversion. -2. Collect *all* rows with ``outcome_samples > 0`` and compute the - category-level *peer_rate* (sample-weighted pooled rate: - Σoutcome_samples × outcome_score / Σoutcome_samples) and - *peer_benchmark* (mean benchmark_score over the same set). + 2. Collect *all* rows with ``outcome_samples > 0`` and compute the + category-level *peer_rate* (sample-weighted pooled rate: + Σoutcome_samples × outcome_score / Σoutcome_samples) and + *peer_benchmark* (mean benchmark_score over the same set). 3. Call ``expected_success_rate`` per row with the benchmark, outcome, and peer statistics. Write the final ``blended_score`` and ``source``. """