Guardrails¶
Unified guardrails suite for temporal validation patterns.
Overview¶
The guardrails module provides a convenience layer over validation gates, aggregating multiple checks into a single validation pass.
Core Classes¶
GuardrailResult¶
Result from guardrail validation:
@dataclass
class GuardrailResult:
passed: bool # All guardrails passed
errors: List[str] # List of error messages
warnings: List[str] # List of warning messages
details: Dict[str, Any] # Additional diagnostic info
Functions¶
run_all_guardrails¶
Run comprehensive validation in one call:
from temporalcv.guardrails import run_all_guardrails
result = run_all_guardrails(
model_metric=0.15,
baseline_metric=0.20,
n_samples=100,
)
if not result.passed:
print(f"Guardrails failed: {result.errors}")
Best Practices¶
Run guardrails before deployment - Catch common issues early
Check both errors and warnings - Warnings indicate edge cases
Use gates for fine-grained control - Guardrails are for convenience