Fix AssertionError crash in MSD binary search when all alphas are feasible

When the network has excess capacity and every probed alpha up to alpha_max
is feasible, the bracketing phase sets upper = min(alpha_max, lower + 1.0).
Since lower already equals alpha_max at that point, upper == lower, causing
the assertion `lower < upper` to fail with an AssertionError.

Fix: return lower directly when no infeasible upper bound is found, since
the last proven feasible alpha is the best answer available.

https://claude.ai/code/session_013XZpjpJ1MM2jJFXHRdEZVi

Author: claude

ngraph/workflow/maximum_supported_demand_step.py

@@ -229,7 +229,9 @@ def _binary_search(self, probe: "Any") -> float:
                     break
                 lower = alpha
             if upper is None:
-                upper = min(self.alpha_max, lower + max(self.resolution, 1.0))
+                # All probed alphas up to alpha_max are feasible;
+                # no infeasible upper bound exists to bisect toward.
+                return lower
         else:
             upper = start_alpha
             alpha = start_alpha

tests/workflow/test_maximum_supported_demand.py

@@ -228,3 +228,42 @@ def test_msd_auto_vs_one_equivalence_single_link() -> None:
     alpha_one = float(exported["steps"]["msd_one"]["data"]["alpha_star"])
     # Both should find approximately the same alpha* for this simple case
     assert abs(alpha_auto - alpha_one) <= 0.02
+
+
+@patch.object(MaximumSupportedDemand, "_evaluate_alpha")
+@patch.object(MaximumSupportedDemand, "_build_cache")
+def test_msd_all_feasible_up_to_alpha_max(
+    mock_build_cache: MagicMock, mock_eval: MagicMock
+) -> None:
+    """Test that MSD returns alpha_max when all alphas are feasible.
+
+    Regression test: when the network has excess capacity and every probed
+    alpha up to alpha_max is feasible, the binary search must return
+    alpha_max instead of crashing with an AssertionError.
+    """
+    mock_build_cache.return_value = MagicMock()
+
+    # Always feasible — network has more capacity than any demand multiplier
+    def _eval(cache, alpha, seeds):
+        return True, {"seeds": 1, "feasible_seeds": 1, "min_placement_ratio": 1.0}
+
+    mock_eval.side_effect = _eval
+
+    scenario = _mock_scenario_with_matrix()
+    alpha_max = 100.0
+    step = MaximumSupportedDemand(
+        name="msd_all_feasible",
+        demand_set="default",
+        alpha_start=1.0,
+        growth_factor=2.0,
+        resolution=0.01,
+        alpha_max=alpha_max,
+        max_bracket_iters=16,
+        seeds_per_alpha=1,
+    )
+    scenario.results = Results()
+    step.execute(scenario)
+
+    exported = scenario.results.to_dict()
+    alpha_star = exported["steps"]["msd_all_feasible"]["data"]["alpha_star"]
+    assert alpha_star == alpha_max