Improve speaker-review excerpts

tests/test_meeting_service.py

@@ -247,6 +247,18 @@ def write_speaker_review_artifacts(run_dir: Path) -> Path:
     return transcript_path
 
 
+def write_custom_speaker_review_artifacts(
+    run_dir: Path, segments: list[dict[str, str | None]]
+) -> Path:
+    """Write custom diarized segments using the standard review context fixture."""
+    transcript_path = write_speaker_review_artifacts(run_dir)
+    transcript_path.write_text(
+        json.dumps({"speaker_labels_anonymous": True, "segments": segments}),
+        encoding="utf-8",
+    )
+    return transcript_path
+
+
 def test_build_context_uses_actual_v1_shape(tmp_path: Path) -> None:
     service, gateway = make_service(tmp_path)
 
@@ -534,6 +546,89 @@ def test_speaker_review_lists_detected_labels_participants_and_excerpts(
     assert "SPEAKER_00" in source.read_text(encoding="utf-8")
 
 
+def test_speaker_review_prefers_content_rich_excerpts_over_acknowledgements_and_questions(
+    tmp_path: Path,
+) -> None:
+    service, gateway = make_service(tmp_path)
+    acknowledgement = "Okay, that sounds good."
+    question = "Could we continue now?"
+    rich_excerpts = [
+        "I will complete the Atlas deployment in Berlin on 14 September with the database team.",
+        "I recommend moving the Orion product launch because the supplier test is incomplete.",
+        "We decided that I will own the Kubernetes migration for the payments service.",
+        "My experience with the Frankfurt factory rollout suggests a two-week validation period.",
+        "I will present the security audit results to the Apollo project steering group.",
+        "We should reserve twelve hours for the API integration and production monitoring.",
+    ]
+    write_custom_speaker_review_artifacts(
+        gateway.run_dir,
+        [{"speaker_id": "SPEAKER_00", "text": text} for text in [acknowledgement, question, *rich_excerpts]],
+    )
+
+    review = service.load_speaker_mapping_review(gateway.run_dir)
+
+    assert review is not None
+    excerpts = review.speakers[0].excerpts
+    assert len(excerpts) == 6
+    assert acknowledgement not in excerpts
+    assert question not in excerpts
+    assert set(excerpts) == set(rich_excerpts)
+
+
+def test_speaker_review_avoids_near_duplicate_excerpts(tmp_path: Path) -> None:
+    service, gateway = make_service(tmp_path)
+    primary = "I will lead the Atlas API migration for the Berlin platform team next Monday."
+    duplicate = "I will lead the Atlas API migration for the Berlin platform group next Monday."
+    alternatives = [
+        "I recommend documenting the supplier quality decision in the project register.",
+        "My team will complete the production calibration before the factory trial.",
+        "I will present the customer research findings at the Munich planning workshop.",
+        "Our database monitoring threshold needs approval from the operations group.",
+        "I will coordinate the Orion release checklist with the support organization.",
+    ]
+    write_custom_speaker_review_artifacts(
+        gateway.run_dir,
+        [{"speaker_id": "SPEAKER_00", "text": text} for text in [primary, duplicate, *alternatives]],
+    )
+
+    review = service.load_speaker_mapping_review(gateway.run_dir)
+
+    assert review is not None
+    excerpts = review.speakers[0].excerpts
+    assert primary in excerpts
+    assert duplicate not in excerpts
+    assert set(excerpts) == {primary, *alternatives}
+
+
+def test_speaker_review_spreads_selected_excerpts_across_meeting_positions(
+    tmp_path: Path,
+) -> None:
+    service, gateway = make_service(tmp_path)
+    positions = {
+        0: "I will lead the Atlas deployment review with the engineering team marker alpha.",
+        1: "I will lead the Atlas deployment review with the engineering team marker bravo.",
+        6: "I will lead the Atlas deployment review with the engineering team marker charlie.",
+        7: "I will lead the Atlas deployment review with the engineering team marker delta.",
+        12: "I will lead the Atlas deployment review with the engineering team marker echo.",
+        13: "I will lead the Atlas deployment review with the engineering team marker foxtrot.",
+        18: "I will lead the Atlas deployment review with the engineering team marker golf.",
+        19: "I will lead the Atlas deployment review with the engineering team marker hotel.",
+    }
+    segments = [
+        {
+            "speaker_id": "SPEAKER_00" if index in positions else "SPEAKER_01",
+            "text": positions.get(index, "Acknowledged."),
+        }
+        for index in range(20)
+    ]
+    write_custom_speaker_review_artifacts(gateway.run_dir, segments)
+
+    review = service.load_speaker_mapping_review(gateway.run_dir)
+
+    assert review is not None
+    assert review.speakers[0].excerpts == tuple(positions[index] for index in (0, 1, 6, 7, 12, 18))
+
+
 def test_protocol_only_regeneration_persists_mappings_without_rewriting_transcript(
     tmp_path: Path,
 ) -> None: