Rename thresholded metric to f1_at_0_75; drop accuracy@0.5 from summaries.

README.md

@@ -80,9 +80,9 @@ No gold-aware snapping. Report **both** layers:
 | **Per event** | **IoU** | Overlap / union of one gold interval vs its matched prediction (within-label 1:1 assignment; unmatched → 0) |
 | **Run / split** | **mean IoU** | Mean of those per-event IoUs |
 | **Run / split** | **continuous F1** | Same number under this protocol: when the model returns the requested multiplicities, \(N_g = N_p\), so continuous F1 \(= 2S/(N_g+N_p)\) collapses to mean IoU. Emitted as `continuous_f1` (= `mean_iou`) for consistency with free-segmentation continuous F1 |
-| **Run / split** | **accuracy@0.5 / @0.75** | Fraction of gold events with IoU ≥ threshold |
+| **Run / split** | **F1@0.75** | Fraction of gold events with IoU ≥ 0.75. Under 1:1 binding with matched counts this equals thresholded precision = recall = F1 — the same F1@0.75 Macrodata uses for free segmentation (here without snap) |
 
-**Takeaway:** IoU is the event-level unit; continuous F1 is the run-level summary of those IoUs (not a different scoring rule). Always quote both names when reporting, and keep accuracy@0.75 as the strict headline next to them.
+**Takeaway:** IoU is the event-level unit; continuous F1 is the run-level summary of those IoUs. F1@0.75 is the Macrodata-aligned publish-style bar on the same bindings. Always quote continuous F1 and F1@0.75 together when reporting.
 
 Shipped code (this repo):
 

localization/score.py

@@ -193,7 +193,6 @@ def score_episode(
                     "pred_start_sec": None if pred is None else pred.start_sec,
                     "pred_end_sec": None if pred is None else pred.end_sec,
                     "iou": iou,
-                    "hit_0_5": iou >= 0.5,
                     "hit_0_75": iou >= 0.75,
                 }
             )
@@ -220,16 +219,16 @@ def metrics_from_rows(rows: Sequence[dict[str, Any]]) -> dict[str, Any]:
     - ``continuous_f1`` is the same number under this protocol (Ng = Np when
       the model returns the requested multiplicities): bound mean IoU ≡
       continuous F1 under one-to-one binding. Both keys are always emitted.
-    - ``accuracy_0_5`` / ``accuracy_0_75``: fraction of gold events with
-      IoU ≥ threshold.
+    - ``f1_at_0_75``: fraction of gold events with IoU ≥ 0.75. Under 1:1
+      binding with Ng = Np this equals thresholded precision = recall = F1,
+      i.e. the same F1@0.75 Macrodata uses for free segmentation (no snap).
     """
     if not rows:
         return {
             "events": 0,
             "mean_iou": None,
             "continuous_f1": None,
-            "accuracy_0_5": None,
-            "accuracy_0_75": None,
+            "f1_at_0_75": None,
         }
     ious = [float(row["iou"]) for row in rows]
     mean_iou = sum(ious) / len(ious)
@@ -237,8 +236,7 @@ def metrics_from_rows(rows: Sequence[dict[str, Any]]) -> dict[str, Any]:
         "events": len(rows),
         "mean_iou": mean_iou,
         "continuous_f1": mean_iou,
-        "accuracy_0_5": sum(iou >= 0.5 for iou in ious) / len(ious),
-        "accuracy_0_75": sum(iou >= 0.75 for iou in ious) / len(ious),
+        "f1_at_0_75": sum(iou >= 0.75 for iou in ious) / len(ious),
     }