diff --git a/src/AcDream.Core/Physics/TransitionTypes.cs b/src/AcDream.Core/Physics/TransitionTypes.cs
index 048d2c4..08c8050 100644
--- a/src/AcDream.Core/Physics/TransitionTypes.cs
+++ b/src/AcDream.Core/Physics/TransitionTypes.cs
@@ -2498,8 +2498,29 @@ public sealed class Transition
                 return true;
 
             // Placement validation: can we actually stand here?
+            //
+            // A6.P3 slice 4 (2026-05-22) — reset WalkInterp to 1.0 before
+            // the placement_insert. The prior TransitionalInsert(5) probe
+            // above may have consumed WalkInterp down to 0 (e.g. on a
+            // step-up onto a sloped ramp — AdjustSphereToPlane lifted the
+            // sphere by step_down_amt and ate all the interp). With
+            // WalkInterp=0, subsequent AdjustSphereToPlane calls' `interp
+            // >= path.WalkInterp` check fires immediately (0 >= 0) and
+            // rejects any push-back — so any geometric overlap that needs
+            // push-out resolution returns false → placement fails → step-up
+            // returns failure → step_up_slide loop → player stuck.
+            //
+            // Retail's CTransition::step_down (acclient_2013_pseudo_c.txt:272952)
+            // resets walk_interp = 1 at function entry. The placement_insert
+            // (line 272989-272990) runs after that initial reset; we believe
+            // placement-insert mode doesn't consume walk_interp the same way
+            // because it's a "can we fit here" check, not a movement sweep.
+            //
+            // This fix is the cellar-up target (issue #98). May also help
+            // other "step-up onto sloped surface" scenarios.
             var savedInsert = sp.InsertType;
             sp.InsertType   = InsertType.Placement;
+            sp.WalkInterp   = 1.0f;
 
             var placeState = TransitionalInsert(1, engine);