feat: add compression receipt drop summaries

app.py

@@ -248,10 +248,15 @@ def _safe_classifier_drop_ranges(
     *,
     text_length: int,
     min_score: float,
-) -> tuple[list[tuple[int, int]], int, int, int]:
+) -> tuple[list[tuple[int, int]], int, int, dict[str, int]]:
+    blocked_by_reason = {
+        "malformed_or_missing_offsets": 0,
+        "out_of_bounds": 0,
+        "below_min_score": 0,
+        "protected_span_overlap": 0,
+    }
     ranges: list[tuple[int, int]] = []
     drop_labels = 0
-    blocked = 0
     for item in labels:
         raw_drop_score = item.get("drop_score")
         try:
@@ -269,19 +274,20 @@ def _safe_classifier_drop_ranges(
             end = int(item["end"])
             score = drop_score if drop_score is not None else float(item.get("score", 1.0))
         except Exception:
-            blocked += 1
+            blocked_by_reason["malformed_or_missing_offsets"] += 1
             continue
-        if (
-            start < 0
-            or end > text_length
-            or end <= start
-            or score < min_score
-            or _overlaps(protected, start, end)
-        ):
-            blocked += 1
+        if start < 0 or end > text_length or end <= start:
+            blocked_by_reason["out_of_bounds"] += 1
+            continue
+        if score < min_score:
+            blocked_by_reason["below_min_score"] += 1
+            continue
+        if _overlaps(protected, start, end):
+            blocked_by_reason["protected_span_overlap"] += 1
             continue
         ranges.append((start, end))
-    return _merge(ranges), drop_labels, len(ranges), blocked
+    merged = _merge(ranges)
+    return merged, drop_labels, len(merged), blocked_by_reason
 
 
 def _is_subsequence(candidate: str, original: str) -> bool:
@@ -368,7 +374,7 @@ def _compress_text(payload: dict[str, Any]) -> dict[str, Any]:
         preserve_json=preserve_json,
     )
     classifier_status, classifier_labels, classifier_error = _manifest_drop_labels(text)
-    classifier_ranges, classifier_drop_labels, classifier_applied, classifier_blocked = (
+    classifier_ranges, classifier_drop_labels, classifier_applied, classifier_blocked_by_reason = (
         _safe_classifier_drop_ranges(
             classifier_labels,
             protected,
@@ -431,6 +437,7 @@ def _compress_text(payload: dict[str, Any]) -> dict[str, Any]:
         {"reason": reason, "preview": preview, "start": start, "end": end}
         for start, end, reason, preview in drops[:20]
     ]
+    classifier_blocked = sum(classifier_blocked_by_reason.values())
     receipt = {
         "protected_spans_checked": len(protected_values),
         "protected_spans_missing": len(missing),
@@ -455,7 +462,14 @@ def _compress_text(payload: dict[str, Any]) -> dict[str, Any]:
             "drop_labels": classifier_drop_labels,
             "drop_spans_applied": classifier_applied,
             "drop_spans_blocked_by_safety": classifier_blocked,
+            "drop_spans_blocked_by_reason": classifier_blocked_by_reason,
         },
+        "removed_spans_count": len(drop_ranges),
+        "removed_chars": sum(end - start for start, end in drop_ranges),
+        "rule_drop_spans_applied": len(drops),
+        "rule_drop_chars": sum(end - start for start, end, _, _ in drops),
+        "classifier_drop_spans_applied": len(classifier_ranges),
+        "classifier_drop_chars": sum(end - start for start, end in classifier_ranges),
         "dropped_segments_count": len(drops),
         "dropped_segments": dropped_segments,
     }