diff --git a/src/promptforge/testing/results.py b/src/promptforge/testing/results.py new file mode 100644 index 0000000..0b477dd --- /dev/null +++ b/src/promptforge/testing/results.py @@ -0,0 +1,33 @@ +from dataclasses import dataclass +from typing import Dict, List, Optional + + +@dataclass +class TestResult: + success: bool + response: str + metrics: "TestMetrics" + error: Optional[str] = None + + +@dataclass +class ComparisonResult: + prompt_name: str + total_runs: int + successful_runs: int + failed_runs: int + avg_latency_ms: float + min_latency_ms: float + max_latency_ms: float + avg_tokens: float + avg_cost: float + success_rate: float + all_metrics: List["TestMetrics"] + + +@dataclass +class TestReport: + test_id: str + timestamp: str + results: Dict[str, ComparisonResult] + summary: Dict[str, float]