Files
audita/tests/test_pipeline.py

97 lines
2.6 KiB
Python

import json
from audita.config import AuditaConfig
from audita.pipeline import process_transcript
from audita.schemas import CorrectionCandidate, CorrectionSet, parse_glossary_yaml, parse_transcript_json
class FakeLLMClient:
def __init__(self, responses):
self.responses = list(responses)
self.calls = 0
def create_corrections(self, messages, config):
self.calls += 1
return self.responses.pop(0)
def _config(tmp_path):
return AuditaConfig(
api_key="key",
max_section_tokens=16000,
confidence_threshold=0.8,
max_retries=3,
work_dir=tmp_path / "work",
)
def _glossary():
return parse_glossary_yaml(
"""
glossary:
- name: "Chauntea"
category: deity
summary: "Chauntea is a deity."
"""
)
def _transcript():
return parse_transcript_json(
"""
[
{"speaker": "Eric", "start": 0.0, "end": 1.0, "text": "I ask Chontia."}
]
"""
)
def test_pipeline_processes_with_fake_llm_and_cleans_work_dir(tmp_path):
correction = CorrectionCandidate(
segment_id=0,
original_text="Chontia",
corrected_text="Chauntea",
confidence=0.95,
)
fake_client = FakeLLMClient([CorrectionSet(corrections=[correction])])
revised = process_transcript(
_transcript(),
_glossary(),
_config(tmp_path),
llm_client=fake_client,
)
assert revised[0].text == "I ask Chauntea."
assert fake_client.calls == 1
assert list((tmp_path / "work").iterdir()) == []
def test_pipeline_skips_bad_correction_and_preserves_diagnostics(tmp_path):
correction = CorrectionCandidate(
segment_id=0,
original_text="Different text.",
corrected_text="Chauntea",
confidence=0.95,
)
fake_client = FakeLLMClient([CorrectionSet(corrections=[correction])])
progress = []
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
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 "does not match any substring" in diagnostics["skipped_corrections"][0]["reason"]