feat(evidence): evaluate retrieval with a small fixture

2026-08-25 01:44:12 +02:00
parent dcb5acc312
commit 619ac2e141
15 changed files with 782 additions and 6 deletions
@@ -115,3 +115,35 @@ def test_evidence_validate_json_is_pristine_and_reports_review_required(monkeypa
        "schemaVersion": 1,
        "status": "review_required",
    }
+
+
+def test_evidence_evaluate_json_reports_the_read_only_generation(monkeypatch, tmp_path):
+    from tht.cli import evidence_cmd
+
+    monkeypatch.setattr(evidence_cmd, "_canonical_worktree", lambda root: root)
+    monkeypatch.setattr(evidence_cmd, "evaluate_from_config", lambda *args, **kwargs: {
+        "workspaceRevision": "a" * 40,
+        "vectorGeneration": "gen:" + "1" * 32,
+        "rrf": {"algorithm": "rrf", "k": 60, "prefetch_limit_multiplier": 2},
+        "passed": True,
+        "countsByExpectedKind": {"formula": 1},
+        "queries": [],
+    })
+
+    result = CliRunner().invoke(app, [
+        "evidence", "evaluate", str(tmp_path), "--json", "-c", str(tmp_path / "runtime.yaml"),
+    ])
+
+    assert result.exit_code == 0
+    assert result.stderr == ""
+    assert json.loads(result.stdout) == {
+        "countsByExpectedKind": {"formula": 1},
+        "operation": "evidence_evaluate",
+        "passed": True,
+        "queries": [],
+        "rrf": {"algorithm": "rrf", "k": 60, "prefetch_limit_multiplier": 2},
+        "schemaVersion": 1,
+        "status": "passed",
+        "vectorGeneration": "gen:" + "1" * 32,
+        "workspaceRevision": "a" * 40,
+    }