Simplified the json schema passed to the LLM, and implemented more forgiving error handling for LLM proposed corrections

tests/test_pipeline.py

@@ -1,9 +1,6 @@
-from pathlib import Path
-
-import pytest
+import json
 
 from audita.config import AuditaConfig
-from audita.errors import AuditaValidationError
 from audita.pipeline import process_transcript
 from audita.schemas import CorrectionCandidate, CorrectionSet, parse_glossary_yaml, parse_transcript_json
 
@@ -51,10 +48,7 @@ def _transcript():
 
 def test_pipeline_processes_with_fake_llm_and_cleans_work_dir(tmp_path):
     correction = CorrectionCandidate(
-        segment_index=0,
-        speaker="Eric",
-        start=0.0,
-        end=1.0,
+        segment_id=0,
         original_text="I ask Chontia.",
         corrected_text="I ask Chauntea.",
         confidence=0.95,
@@ -73,27 +67,30 @@ def test_pipeline_processes_with_fake_llm_and_cleans_work_dir(tmp_path):
     assert list((tmp_path / "work").iterdir()) == []
 
 
-def test_pipeline_preserves_work_dir_on_failure(tmp_path):
+def test_pipeline_skips_bad_correction_and_preserves_diagnostics(tmp_path):
     correction = CorrectionCandidate(
-        segment_index=0,
-        speaker="Eric",
-        start=0.0,
-        end=1.0,
+        segment_id=0,
         original_text="Different text.",
         corrected_text="I ask Chauntea.",
         confidence=0.95,
     )
     fake_client = FakeLLMClient([CorrectionSet(corrections=[correction])])
+    progress = []
 
-    with pytest.raises(AuditaValidationError):
-        process_transcript(
-            _transcript(),
-            _glossary(),
-            _config(tmp_path),
-            llm_client=fake_client,
-        )
+    revised = process_transcript(
+        _transcript(),
+        _glossary(),
+        _config(tmp_path),
+        llm_client=fake_client,
+        progress=progress.append,
+    )
 
+    assert revised[0].text == "I ask Chontia."
+    assert any("Skipping correction for segment 0" in message for message in progress)
     preserved = list((tmp_path / "work").iterdir())
     assert len(preserved) == 1
-    assert (Path(preserved[0]) / "section-0000.json").exists()
-
+    skipped_path = preserved[0] / "skipped-corrections.json"
+    assert skipped_path.exists()
+    diagnostics = json.loads(skipped_path.read_text(encoding="utf-8"))
+    assert diagnostics["skipped_corrections"][0]["segment_id"] == 0
+    assert "original_text" in diagnostics["skipped_corrections"][0]["reason"]