Implement support for segments with an optional categories key

tests/test_new_pipeline.py

@@ -46,9 +46,9 @@ def _transcript():
     return parse_source_transcript_json(
         """
         [
-          {"speaker": "Eric", "start": 0.0, "end": 1.0, "text": "Hello."},
-          {"speaker": "Eric", "start": 1.5, "end": 2.0, "text": "Again."},
-          {"speaker": "Mike", "start": 10.0, "end": 11.0, "text": "Done."}
+          {"speaker": "Eric", "start": 0.0, "end": 1.0, "text": "Hello.", "categories": ["intro"]},
+          {"speaker": "Eric", "start": 1.5, "end": 2.0, "text": "Again.", "categories": ["intro", "aside"]},
+          {"speaker": "Mike", "start": 10.0, "end": 11.0, "text": "Done.", "categories": ["response"]}
         ]
         """
     )
@@ -74,6 +74,8 @@ def test_process_transcript_runs_noop_framework(tmp_path):
     assert [segment.id for segment in revised] == [1, 2]
     assert revised[0].text == "Hello. Again."
     assert revised[1].text == "Done."
+    assert revised[0].categories == ["intro", "aside"]
+    assert revised[1].categories == ["response"]
     assert [call["stage_name"] for call in llm_client.calls] == [
         "glossary_1:proposal",
         "homophones:proposal",
@@ -186,6 +188,30 @@ def test_external_report_can_be_written(tmp_path):
     assert payload["totals"]["applied_change_count"] == 0
 
 
+def test_process_transcript_preserves_categories_in_llm_prompt_payloads(tmp_path):
+    llm_client = FakeStructuredLLMClient(
+        [
+            {"corrections": []},
+            {"corrections": []},
+            {"corrections": []},
+            {"corrections": []},
+            {"corrections": []},
+        ]
+    )
+
+    process_transcript(
+        _transcript(),
+        _glossary(),
+        AuditaConfig.from_sources(env={}, overrides=None),
+        llm_client=llm_client,
+    )
+
+    proposal_prompt = llm_client.calls[0]["messages"][1]["content"]
+    assert '"categories": [' in proposal_prompt
+    assert '"intro"' in proposal_prompt
+    assert '"aside"' in proposal_prompt
+
+
 def test_default_module_specs_expose_final_validator_order():
     specs = default_module_specs()