Implemented optional concurrency for the LLM backend

tests/test_framework_runner.py

@@ -1,4 +1,7 @@
-from audita.core.config import AuditaConfig
+import threading
+
+from audita.core.chunking import IndexedSegment, TranscriptSection
+from audita.core.config import AuditaConfig, ConfigOverrides
 from audita.core.errors import AuditaLLMError
 from audita.core.schemas import parse_glossary_yaml, parse_transcript_json
 from audita.framework.models import CorrectionProposal, ModuleContext, ModuleRunSpec
@@ -76,6 +79,35 @@ class RecordingModule:
         return list(self._proposals)
 
 
+class ConcurrentRecordingModule:
+    replacement_policy = "require_unique"
+
+    def __init__(self, recorder, barrier):
+        self.module_key = "concurrent"
+        self._recorder = recorder
+        self._barrier = barrier
+
+    def validators(self):
+        return []
+
+    def propose(self, transcript_section, context: ModuleContext):
+        texts = [item.segment.text for item in transcript_section.segments]
+        self._recorder.append(("start", transcript_section.section_index, texts))
+        self._barrier.wait()
+        segment = transcript_section.segments[0].segment
+        return [
+            CorrectionProposal(
+                proposal_index=0,
+                module_instance=context.run_spec.instance_name,
+                module_key=context.run_spec.module_key,
+                id=segment.id,
+                original_text=segment.text.rstrip("."),
+                corrected_text=f"{segment.text.rstrip('.')} revised",
+                confidence=0.9,
+            )
+        ]
+
+
 def test_pipeline_runner_applies_modules_sequentially(tmp_path):
     transcript = parse_transcript_json(
         """
@@ -382,3 +414,54 @@ def test_pipeline_runner_uses_real_protected_glossary_validator(tmp_path):
     assert result.transcript[0].text == "Hrank moves."
     assert result.skipped_corrections[0].source == "validator:protected_glossary_guard"
     assert result.skipped_corrections[0].reason == "correction changes protected glossary term usage"
+
+
+def test_pipeline_runner_collects_section_proposals_concurrently_and_preserves_section_order(tmp_path, monkeypatch):
+    transcript = parse_transcript_json(
+        """
+        [
+          {"id": 1, "speaker": "A", "start": 0.0, "end": 1.0, "text": "Alpha."},
+          {"id": 2, "speaker": "A", "start": 1.0, "end": 2.0, "text": "Beta."}
+        ]
+        """
+    )
+    glossary = parse_glossary_yaml(
+        """
+        glossary:
+          - name: "Alpha"
+            category: noun
+            summary: "Alpha."
+        """
+    )
+    sections = [
+        TranscriptSection(
+            section_index=0,
+            start_index=0,
+            segments=[IndexedSegment(index=0, segment=transcript[0])],
+            token_count=1,
+        ),
+        TranscriptSection(
+            section_index=1,
+            start_index=1,
+            segments=[IndexedSegment(index=1, segment=transcript[1])],
+            token_count=1,
+        ),
+    ]
+    seen = []
+    module = ConcurrentRecordingModule(seen, threading.Barrier(2, timeout=1.0))
+    monkeypatch.setattr("audita.framework.runner.chunk_transcript", lambda working, max_tokens: sections)
+
+    runner = PipelineRunner()
+    result = runner.run(
+        transcript=transcript,
+        glossary=glossary,
+        module_specs=[ModuleRunSpec(instance_name="concurrent", module_key="concurrent", module=module)],
+        config=AuditaConfig.from_sources(env={}, overrides=ConfigOverrides(llm_concurrency=2)),
+        run_dir=tmp_path / "run",
+    )
+
+    assert [change.id for change in result.applied_changes] == [1, 2]
+    assert result.applied_changes[0].corrected_text == "Alpha revised"
+    assert result.applied_changes[1].corrected_text == "Beta revised"
+    assert result.transcript[0].text == "Alpha revised."
+    assert result.transcript[1].text == "Beta revised."