diff --git a/subwave_nowplaying.star b/subwave_nowplaying.star
index 77e6caa..9fb94ef 100644
--- a/subwave_nowplaying.star
+++ b/subwave_nowplaying.star
@@ -400,16 +400,22 @@ GLITCH_STATE_POOL = [
     # two scanline rows; accepted, since the shorter height was the ask.
     #
     # hue2 (-60.0, magenta) is placed 3-5px *above* hue1 (150.0, cyan) in
-    # every entry (top2 = top1 - a 3-5px gap) -- a fixed relative offset
-    # rather than independently-random positions, per request.
-    (10, 3, -2, 150.0, 6, 4, 2, -60.0),
-    (14, 4, 2, 150.0, 11, 3, -3, -60.0),
-    (8, 2, -3, 150.0, 3, 4, 1, -60.0),
-    (20, 4, 1, 150.0, 16, 2, -2, -60.0),
-    (6, 3, -1, 150.0, 3, 3, 3, -60.0),
-    (24, 2, 3, 150.0, 19, 4, -1, -60.0),
-    (12, 3, -2, 150.0, 8, 2, 1, -60.0),
-    (18, 4, 1, 150.0, 15, 3, -3, -60.0),
+    # every entry -- a fixed relative offset rather than
+    # independently-random positions, per request. The gap is measured
+    # edge-to-edge (top1 - (top2 + h2)), not top-to-top (top1 - top2):
+    # an earlier version used top-to-top, which let band2 (magenta)
+    # overlap band1 (cyan) whenever h2 exceeded the top-to-top gap (e.g.
+    # h2=4 with a 3px top-to-top gap put magenta's bottom edge 1px into
+    # cyan's territory). top2 = top1 - h2 - gap guarantees a real 3-5px
+    # of clean space between them regardless of h2.
+    (12, 3, -2, 150.0, 4, 4, 2, -60.0),
+    (16, 4, 2, 150.0, 10, 3, -3, -60.0),
+    (14, 2, -3, 150.0, 5, 4, 1, -60.0),
+    (22, 4, 1, 150.0, 16, 2, -2, -60.0),
+    (10, 3, -1, 150.0, 3, 3, 3, -60.0),
+    (26, 2, 3, 150.0, 19, 4, -1, -60.0),
+    (14, 3, -2, 150.0, 8, 2, 1, -60.0),
+    (20, 4, 1, 150.0, 14, 3, -3, -60.0),
 ]
 
 # render.Sequence doesn't loop on its own (see _glitch_scanlines below), so