Implemented a validation step to confirm that grammatical changes do not change substantive meaning

This commit is contained in:
2026-04-22 15:58:05 -05:00
parent 87c47517a0
commit d7788da1f4
15 changed files with 849 additions and 8 deletions

View File

@@ -8,6 +8,8 @@ from audita.config import (
DEFAULT_GLOSSARY_MAX_LLM_PASSES,
DEFAULT_GRAMMAR_CONFIDENCE_THRESHOLD,
DEFAULT_GRAMMAR_MAX_LLM_PASSES,
DEFAULT_GRAMMAR_VALIDATION_CONFIDENCE_THRESHOLD,
DEFAULT_GRAMMAR_VALIDATION_ENABLED,
DEFAULT_MAX_RETRIES,
DEFAULT_MAX_SECTION_TOKENS,
DEFAULT_NORMALIZE_ELLIPSIS_GAP,
@@ -30,6 +32,10 @@ def test_config_uses_defaults_with_api_key():
assert config.max_retries == DEFAULT_MAX_RETRIES
assert config.glossary_max_llm_passes == DEFAULT_GLOSSARY_MAX_LLM_PASSES
assert config.grammar_max_llm_passes == DEFAULT_GRAMMAR_MAX_LLM_PASSES
assert config.grammar_validation_enabled == DEFAULT_GRAMMAR_VALIDATION_ENABLED
assert config.grammar_validation_enabled is True
assert config.grammar_validation_confidence_threshold == DEFAULT_GRAMMAR_VALIDATION_CONFIDENCE_THRESHOLD
assert config.grammar_validation_confidence_threshold == 0.8
assert config.normalize_max_segment_gap == DEFAULT_NORMALIZE_MAX_SEGMENT_GAP
assert config.normalize_max_segment_gap == 5.0
assert config.normalize_ellipsis_gap == DEFAULT_NORMALIZE_ELLIPSIS_GAP
@@ -50,6 +56,8 @@ def test_config_env_overrides_defaults():
"AUDITA_MAX_RETRIES": "5",
"AUDITA_GLOSSARY_MAX_LLM_PASSES": "7",
"AUDITA_GRAMMAR_MAX_LLM_PASSES": "4",
"AUDITA_GRAMMAR_VALIDATION_ENABLED": "false",
"AUDITA_GRAMMAR_VALIDATION_CONFIDENCE_THRESHOLD": "0.91",
"AUDITA_NORMALIZE_MAX_SEGMENT_GAP": "4.5",
"AUDITA_NORMALIZE_ELLIPSIS_GAP": "1.5",
"AUDITA_NORMALIZE_MAX_SEGMENT_DURATION": "45.0",
@@ -64,6 +72,8 @@ def test_config_env_overrides_defaults():
assert config.max_retries == 5
assert config.glossary_max_llm_passes == 7
assert config.grammar_max_llm_passes == 4
assert config.grammar_validation_enabled is False
assert config.grammar_validation_confidence_threshold == 0.91
assert config.normalize_max_segment_gap == 4.5
assert config.normalize_ellipsis_gap == 1.5
assert config.normalize_max_segment_duration == 45.0
@@ -79,6 +89,8 @@ def test_config_cli_overrides_env():
"AUDITA_MAX_RETRIES": "5",
"AUDITA_GLOSSARY_MAX_LLM_PASSES": "7",
"AUDITA_GRAMMAR_MAX_LLM_PASSES": "6",
"AUDITA_GRAMMAR_VALIDATION_ENABLED": "false",
"AUDITA_GRAMMAR_VALIDATION_CONFIDENCE_THRESHOLD": "0.91",
"AUDITA_NORMALIZE_MAX_SEGMENT_GAP": "4.5",
"AUDITA_NORMALIZE_ELLIPSIS_GAP": "1.5",
"AUDITA_NORMALIZE_MAX_SEGMENT_DURATION": "45.0",
@@ -92,6 +104,8 @@ def test_config_cli_overrides_env():
max_retries=3,
glossary_max_llm_passes=2,
grammar_max_llm_passes=3,
grammar_validation_enabled=True,
grammar_validation_confidence_threshold=0.75,
normalize_max_segment_gap=3.0,
normalize_ellipsis_gap=1.0,
normalize_max_segment_duration=30.0,
@@ -106,6 +120,8 @@ def test_config_cli_overrides_env():
assert config.max_retries == 3
assert config.glossary_max_llm_passes == 2
assert config.grammar_max_llm_passes == 3
assert config.grammar_validation_enabled is True
assert config.grammar_validation_confidence_threshold == 0.75
assert config.normalize_max_segment_gap == 3.0
assert config.normalize_ellipsis_gap == 1.0
assert config.normalize_max_segment_duration == 30.0
@@ -148,6 +164,17 @@ def test_config_rejects_invalid_stage_thresholds():
AuditaConfig.from_sources(
env={"OPENROUTER_API_KEY": "key", "AUDITA_GRAMMAR_CONFIDENCE_THRESHOLD": "-0.1"}
)
with pytest.raises(AuditaConfigError):
AuditaConfig.from_sources(
env={"OPENROUTER_API_KEY": "key", "AUDITA_GRAMMAR_VALIDATION_CONFIDENCE_THRESHOLD": "1.1"}
)
def test_config_rejects_invalid_grammar_validation_enabled():
with pytest.raises(AuditaConfigError):
AuditaConfig.from_sources(
env={"OPENROUTER_API_KEY": "key", "AUDITA_GRAMMAR_VALIDATION_ENABLED": "maybe"}
)
def test_legacy_confidence_threshold_env_is_ignored():