[kimi] Refactor autoresearch.py -> SystemExperiment class (#906) (#1244)

Co-authored-by: Kimi Agent <kimi@timmy.local>
Co-committed-by: Kimi Agent <kimi@timmy.local>

tests/timmy/test_autoresearch.py

@@ -39,7 +39,9 @@ class TestPrepareExperiment:
         from timmy.autoresearch import prepare_experiment
 
         with patch("timmy.autoresearch.subprocess.run") as mock_run:
-            mock_run.return_value = MagicMock(returncode=1, stdout="", stderr="auth failed")
+            mock_run.return_value = MagicMock(
+                returncode=1, stdout="", stderr="auth failed"
+            )
             result = prepare_experiment(tmp_path)
 
         assert "failed" in result.lower()
@@ -102,7 +104,9 @@ class TestRunExperiment:
         (repo_dir / "train.py").write_text("print('done')")
 
         with patch("timmy.autoresearch.subprocess.run") as mock_run:
-            mock_run.return_value = MagicMock(returncode=0, stdout="no metrics here", stderr="")
+            mock_run.return_value = MagicMock(
+                returncode=0, stdout="no metrics here", stderr=""
+            )
             result = run_experiment(tmp_path)
 
         assert result["success"] is True
@@ -349,3 +353,107 @@ class TestSystemExperiment:
             success = exp.revert_changes()
 
         assert success is False
+
+    def test_create_branch_success(self, tmp_path):
+        from timmy.autoresearch import SystemExperiment
+
+        exp = SystemExperiment(target="x.py", workspace=tmp_path)
+        with patch("timmy.autoresearch.subprocess.run") as mock_run:
+            mock_run.return_value = MagicMock(returncode=0)
+            success = exp.create_branch("feature/test-branch")
+
+        assert success is True
+        # Verify correct git command was called
+        mock_run.assert_called_once()
+        call_args = mock_run.call_args[0][0]
+        assert "checkout" in call_args
+        assert "-b" in call_args
+        assert "feature/test-branch" in call_args
+
+    def test_create_branch_failure(self, tmp_path):
+        import subprocess
+
+        from timmy.autoresearch import SystemExperiment
+
+        exp = SystemExperiment(target="x.py", workspace=tmp_path)
+        with patch("timmy.autoresearch.subprocess.run") as mock_run:
+            mock_run.side_effect = subprocess.CalledProcessError(1, "git")
+            success = exp.create_branch("feature/test-branch")
+
+        assert success is False
+
+    def test_run_dry_run_mode(self, tmp_path):
+        """Test that run() in dry_run mode only generates hypotheses."""
+        from timmy.autoresearch import SystemExperiment
+
+        exp = SystemExperiment(target="x.py", workspace=tmp_path)
+        result = exp.run(max_iterations=3, dry_run=True, program_content="Test program")
+
+        assert result["iterations"] == 3
+        assert result["success"] is False  # No actual experiments run
+        assert len(exp.results) == 3
+        # Each result should have a hypothesis
+        for record in exp.results:
+            assert "hypothesis" in record
+
+    def test_run_with_custom_metric_fn(self, tmp_path):
+        """Test that custom metric_fn is used for metric extraction."""
+        from timmy.autoresearch import SystemExperiment
+
+        def custom_metric_fn(output: str) -> float | None:
+            match = __import__("re").search(r"custom_metric:\s*([0-9.]+)", output)
+            return float(match.group(1)) if match else None
+
+        exp = SystemExperiment(
+            target="x.py",
+            workspace=tmp_path,
+            metric="custom",
+            metric_fn=custom_metric_fn,
+        )
+
+        with patch("timmy.autoresearch.subprocess.run") as mock_run:
+            mock_run.return_value = MagicMock(
+                returncode=0,
+                stdout="custom_metric: 42.5\nother output",
+                stderr="",
+            )
+            tox_result = exp.run_tox()
+
+        assert tox_result["metric"] == pytest.approx(42.5)
+
+    def test_run_single_iteration_success(self, tmp_path):
+        """Test a successful single iteration that finds an improvement."""
+        from timmy.autoresearch import SystemExperiment
+
+        exp = SystemExperiment(target="x.py", workspace=tmp_path)
+
+        with patch("timmy.autoresearch.subprocess.run") as mock_run:
+            # Mock tox returning a passing test with metric
+            mock_run.return_value = MagicMock(
+                returncode=0,
+                stdout="10 passed in 1.23s",
+                stderr="",
+            )
+            result = exp.run(max_iterations=1, tox_env="unit")
+
+        assert result["iterations"] == 1
+        assert len(exp.results) == 1
+        assert exp.results[0]["metric"] == pytest.approx(100.0)
+
+    def test_run_stores_baseline_on_first_success(self, tmp_path):
+        """Test that baseline is set after first successful iteration."""
+        from timmy.autoresearch import SystemExperiment
+
+        exp = SystemExperiment(target="x.py", workspace=tmp_path)
+        assert exp.baseline is None
+
+        with patch("timmy.autoresearch.subprocess.run") as mock_run:
+            mock_run.return_value = MagicMock(
+                returncode=0,
+                stdout="8 passed in 1.23s",
+                stderr="",
+            )
+            exp.run(max_iterations=1)
+
+        assert exp.baseline == pytest.approx(100.0)
+        assert exp.results[0]["baseline"] is None  # First run has no baseline