Replaced the stub LLM-based validator class with two real LLM-based validator implementations

tests/test_framework_runner.py

@@ -1,8 +1,9 @@
 from audita.core.config import AuditaConfig
+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
 from audita.framework.runner import PipelineRunner
-from audita.validators import ProtectedGlossaryTermsValidator
+from audita.validators import MeaningReversalValidator, ProtectedGlossaryTermsValidator, SpokenFormPlausibilityValidator
 from audita.validators.base import ValidationContext, ValidationDecision, ValidationResult
 
 
@@ -34,6 +35,25 @@ class RecordingLLMValidator(RecordingValidator):
     execution_kind = "llm"
 
 
+class FakeStructuredLLMClient:
+    def __init__(self, responses):
+        self._responses = list(responses)
+        self.calls = []
+
+    def run_structured(self, *, stage_name, messages, response_model, config):
+        self.calls.append(
+            {
+                "stage_name": stage_name,
+                "messages": list(messages),
+                "response_model": response_model,
+            }
+        )
+        if not self._responses:
+            raise AuditaLLMError("FakeStructuredLLMClient received more calls than expected.")
+        payload = self._responses.pop(0)
+        return response_model.model_validate(payload)
+
+
 class RecordingModule:
     replacement_policy = "require_unique"
 
@@ -227,6 +247,89 @@ def test_pipeline_runner_supports_deterministic_and_llm_validators_in_one_chain(
     ]
 
 
+def test_pipeline_runner_supports_real_llm_validators_in_one_chain(tmp_path):
+    transcript = parse_transcript_json(
+        """
+        [
+          {"id": 1, "speaker": "A", "start": 0.0, "end": 1.0, "text": "There were gestures at the dam."}
+        ]
+        """
+    )
+    glossary = parse_glossary_yaml(
+        """
+        glossary:
+          - name: "Jesters"
+            category: faction
+            summary: "The Jesters are a faction."
+        """
+    )
+    module = RecordingModule(
+        "mixed_real",
+        [
+            CorrectionProposal(
+                proposal_index=0,
+                module_instance="mixed_real",
+                module_key="glossary",
+                id=1,
+                original_text="gestures",
+                corrected_text="Jesters",
+                confidence=0.9,
+            )
+        ],
+        [
+            ProtectedGlossaryTermsValidator("protected_glossary_guard"),
+            SpokenFormPlausibilityValidator("spoken_form_plausibility_review"),
+            MeaningReversalValidator("meaning_reversal_review"),
+        ],
+        [],
+    )
+    llm_client = FakeStructuredLLMClient(
+        [
+            {
+                "validations": [
+                    {
+                        "correction_index": 0,
+                        "approved": True,
+                        "confidence": 0.95,
+                        "reason": "Likely phonetic mistranscription in context.",
+                    }
+                ]
+            },
+            {
+                "validations": [
+                    {
+                        "correction_index": 0,
+                        "approved": True,
+                        "confidence": 0.98,
+                        "reason": "Does not reverse the segment meaning.",
+                    }
+                ]
+            },
+        ]
+    )
+
+    runner = PipelineRunner()
+    result = runner.run(
+        transcript=transcript,
+        glossary=glossary,
+        module_specs=[ModuleRunSpec(instance_name="mixed_real", module_key="glossary", module=module)],
+        config=AuditaConfig.from_sources(env={"OPENROUTER_API_KEY": "test-key"}),
+        run_dir=tmp_path / "run",
+        llm_client=llm_client,
+    )
+
+    assert result.transcript[0].text == "There were Jesters at the dam."
+    assert [report.execution_kind for report in result.module_reports[0].validators] == [
+        "deterministic",
+        "llm",
+        "llm",
+    ]
+    assert [call["stage_name"] for call in llm_client.calls] == [
+        "mixed_real:spoken_form_plausibility_review",
+        "mixed_real:meaning_reversal_review",
+    ]
+
+
 def test_pipeline_runner_uses_real_protected_glossary_validator(tmp_path):
     transcript = parse_transcript_json(
         """