diff --git a/docs/roadmap.md b/docs/roadmap.md
index 8d4eb13..e05a03d 100644
--- a/docs/roadmap.md
+++ b/docs/roadmap.md
@@ -725,11 +725,18 @@ lookup, and it is the main thing inherited from ReRAC.
       every decoded tfrag texture on 27 levels: **97.94% fully opaque, 0.98%
       fully clear, 1.08% partial across 126 distinct values.** So the `0x80 -> 1.0`
       rule is not a corner case -- it is what 22.7 million texels do -- while
-      blending touches about 1% of them. **Alpha testing (cutout) matters far more
-      than alpha blending on this disc**, and the 126 evenly-populated partial
-      values look like smooth alpha (soft edges, foliage, distance fade) rather
-      than a quantised blend ladder. That reorders the remaining GS work: the
-      cutout path is the one that has to be right.
+      blending touches about 1.08% of them. **These are the same size** -- the cutout
+      path handles the 0.98% fully-clear texels, the blend path the 1.08%
+      partial -- so an earlier version of this note claiming cutout "matters far
+      more" was not supported by the counts it cited. What does make cutout more
+      consequential is the *shape* of the failure: a clear texel drawn when it
+      should be discarded puts geometry where there is none, which is visible,
+      while a partial texel drawn opaque is a slightly wrong soft edge, which is
+      not. Cutout first is a claim about which failure you notice, not about how
+      many texels are involved.
+
+      The 126 evenly-populated partial values look like smooth alpha (soft edges,
+      foliage, distance fade) rather than a quantised blend ladder.
 
       **No draw-path or shader work has started.**
 - [ ] VU0/VU1 skinning and lighting on the PS2 float model