Initial standalone memabra release

tests/test_training_reports.py

@@ -0,0 +1,74 @@
+from __future__ import annotations
+
+from datetime import datetime, timezone
+
+from memabra.evaluator import EvaluationResult
+from memabra.promotion import PromotionDecision, PromotionPolicy
+from memabra.training_reports import TrainingReportStore, build_report
+
+
+def test_build_report_includes_all_required_fields():
+    baseline = EvaluationResult(task_count=2, avg_reward=0.5, error_rate=0.1, avg_latency_ms=50.0)
+    challenger = EvaluationResult(task_count=2, avg_reward=0.6, error_rate=0.05, avg_latency_ms=45.0)
+    decision = PromotionDecision(accepted=True, reasons=[], metrics={"reward_delta": 0.1})
+
+    report = build_report(
+        source_trajectory_ids=["t1", "t2"],
+        baseline=baseline,
+        challenger=challenger,
+        decision=decision,
+        promoted_version_id="v-2026",
+    )
+
+    assert report["source_trajectory_ids"] == ["t1", "t2"]
+    assert report["sample_count"] == 2
+    assert "timestamp" in report
+    assert report["promoted_version_id"] == "v-2026"
+    assert report["baseline_metrics"]["avg_reward"] == 0.5
+    assert report["challenger_metrics"]["avg_reward"] == 0.6
+    assert report["promotion_decision"]["accepted"] is True
+
+
+def test_training_report_store_save_and_list(tmp_path):
+    store = TrainingReportStore(base_dir=tmp_path / "reports")
+    report = build_report(
+        source_trajectory_ids=["t1"],
+        baseline=EvaluationResult(task_count=1, avg_reward=0.5, error_rate=0.0, avg_latency_ms=10.0),
+        challenger=EvaluationResult(task_count=1, avg_reward=0.6, error_rate=0.0, avg_latency_ms=10.0),
+        decision=PromotionDecision(accepted=False, reasons=["reward too low"], metrics={}),
+    )
+
+    saved = store.save(report)
+    reports = store.list_reports()
+
+    assert len(reports) == 1
+    assert reports[0]["report_id"] == saved["report_id"]
+    assert reports[0]["promotion_decision"]["accepted"] is False
+
+
+def test_training_report_store_get_report_returns_specific_report(tmp_path):
+    from memabra.training_reports import TrainingReportStore, build_report
+    from memabra.evaluator import EvaluationResult
+    from memabra.promotion import PromotionDecision
+
+    store = TrainingReportStore(base_dir=tmp_path)
+    report = build_report(
+        source_trajectory_ids=["t1", "t2"],
+        baseline=EvaluationResult(task_count=1, trajectories=[], avg_reward=0.5, error_rate=0.0, avg_latency_ms=10.0, decision_distribution={}),
+        challenger=EvaluationResult(task_count=1, trajectories=[], avg_reward=0.6, error_rate=0.0, avg_latency_ms=10.0, decision_distribution={}),
+        decision=PromotionDecision(accepted=True, reasons=[], metrics={}),
+        promoted_version_id="v1",
+    )
+    store.save(report)
+
+    fetched = store.get_report(report["report_id"])
+    assert fetched is not None
+    assert fetched["report_id"] == report["report_id"]
+    assert fetched["promoted_version_id"] == "v1"
+
+
+def test_training_report_store_get_report_missing_returns_none(tmp_path):
+    from memabra.training_reports import TrainingReportStore
+
+    store = TrainingReportStore(base_dir=tmp_path)
+    assert store.get_report("nonexistent") is None