fix: dedupe conflict findings by document pair

tests/test_conflicts.py

@@ -67,6 +67,23 @@ def test_chunks_from_one_document_are_not_a_conflict_with_themselves():
     assert detect_conflicts([_hit("hr-policy"), _hit("hr-policy")]) == []
 
 
+def test_chunks_from_two_official_documents_are_reported_once():
+    """Retrieval returns chunks, not documents. One disagreement must not be printed once per
+    chunk pair."""
+    hits = [
+        _hit("hr-policy"),
+        _hit("hr-policy"),
+        _hit("finance-policy"),
+        _hit("finance-policy"),
+    ]
+    assert len(detect_conflicts(hits)) == 1
+
+
+def test_official_chunks_do_not_duplicate_an_informal_conflict():
+    hits = [_hit("hr-policy"), _hit("hr-policy"), _hit("notes", official=False)]
+    assert len(detect_conflicts(hits)) == 1
+
+
 def test_a_single_source_of_truth_produces_no_conflict():
     assert detect_conflicts([_hit("hr-policy")]) == []
 

vaultrag/conflicts.py

@@ -49,11 +49,16 @@ def detect_conflicts(hits: list[Hit]) -> list[Conflict]:
     """
     out: list[Conflict] = []
     official = [h for h in hits if h.is_official]
+    seen: set[frozenset[str]] = set()
 
     for i, a in enumerate(official):
         for b in official[i + 1 :]:
             if a.doc_id == b.doc_id:
                 continue
+            pair = frozenset((a.doc_id, b.doc_id))
+            if pair in seen:
+                continue
+            seen.add(pair)
             out.append(
                 Conflict(
                     doc_a=a.doc_id,
@@ -70,6 +75,10 @@ def detect_conflicts(hits: list[Hit]) -> list[Conflict]:
     for a in official:
         for b in unofficial:
             if a.doc_id != b.doc_id:
+                pair = frozenset((a.doc_id, b.doc_id))
+                if pair in seen:
+                    break
+                seen.add(pair)
                 out.append(
                     Conflict(
                         doc_a=a.doc_id,