Add policy_eval.py/tests/test_core.py

policy_eval.py/tests/test_core.py

@@ -0,0 +1,70 @@
+import json
+import os
+import tempfile
+import pytest
+import pandas as pd
+from pathlib import Path
+
+import src.policy_eval as policy_eval
+
+@pytest.fixture
+def policy_constants_sample():
+    return {"version": "1.0.0", "constant_value": 42.0}
+
+@pytest.fixture
+def changed_policy_constants():
+    return {"version": "1.0.1", "constant_value": 43.0}
+
+@pytest.fixture
+def audit_set_tmp(tmp_path):
+    audit_dir = tmp_path / "audit"
+    audit_dir.mkdir()
+    csv_path = audit_dir / "data.csv"
+    df = pd.DataFrame({"id": [1,2], "value": [10,20]})
+    df.to_csv(csv_path, index=False)
+    return str(audit_dir)
+
+def test_check_policy_changes_no_change(policy_constants_sample):
+    # When the same constants are checked twice, should detect no change
+    assert policy_eval.check_policy_changes(policy_constants_sample) == False
+    # Second run with same should still result False (deterministic behavior)
+    assert policy_eval.check_policy_changes(policy_constants_sample) == False
+
+def test_check_policy_changes_detects_change(policy_constants_sample, changed_policy_constants):
+    # First baseline call
+    _ = policy_eval.check_policy_changes(policy_constants_sample)
+    result = policy_eval.check_policy_changes(changed_policy_constants)
+    assert result is True
+
+@pytest.mark.parametrize("variant", ["small", "medium"])
+def test_run_backtest_creates_outputs(audit_set_tmp, variant):
+    results = policy_eval.run_backtest(audit_set_tmp)
+    assert isinstance(results, dict)
+    summary_path = results.get("delta_summary")
+    cases_path = results.get("delta_cases")
+
+    assert summary_path and os.path.isfile(summary_path)
+    assert cases_path and os.path.isfile(cases_path)
+
+    with open(summary_path) as f:
+        summary = json.load(f)
+    assert isinstance(summary, dict)
+
+    df_cases = pd.read_csv(cases_path)
+    expected_cols = ["reason", "old_decision", "new_decision", "policy_hash"]
+    for col in expected_cols:
+        assert col in df_cases.columns
+
+    # determinism: calling again produces same summary
+    second_results = policy_eval.run_backtest(audit_set_tmp)
+    with open(second_results["delta_summary"]) as f2:
+        summary2 = json.load(f2)
+    assert summary == summary2
+
+def test_run_backtest_with_empty_audit(tmp_path):
+    empty_dir = tmp_path / "empty"
+    empty_dir.mkdir()
+    result = policy_eval.run_backtest(str(empty_dir))
+    assert isinstance(result["delta_summary"], str)
+    assert os.path.exists(result["delta_summary"])
+    assert os.path.exists(result["delta_cases"])