@@ -620,7 +620,12 @@ def paint(pal, u0, u1, v0, v1):
     # block on the KOS-MOS atlas measures below that) but the canvas read
     # directly as CT32 is more coherent, draw the block from the canvas —
     # each canvas pixel covers 2x2 output pixels.
-    for v0 in range(0, H, 64):
+    # Only for model-bound textures (a .lex exists): true-colour regions are
+    # a character-model feature, and on the lex-less UI sheets the pass
+    # mis-fired on dithered 8bpp art — the casino slot reels (two indices
+    # alternating) read as a "coherent" checker when taken as CT32 and were
+    # repainted blue/yellow (regression vs the July build, found 2026-09-24).
+    for v0 in (range(0, H, 64) if lexes else ()):
         v1 = min(H, v0 + 64)
         for u0 in range(0, W, 64):
             u1 = min(W, u0 + 64)
 
@@ -185,7 +185,11 @@ still measures as dither noise (`_rgba_noise` ratio >= 0.9; correctly
 paletted blocks measure below that) and the canvas region read directly
 as CT32 is more coherent, the block is drawn straight from the canvas
 (each canvas pixel = 2x2 output pixels, alpha scaled 7->8 bit). This
-finally renders the KOS-MOS hair band as hair instead of noise.
+finally renders the KOS-MOS hair band as hair instead of noise. The pass
+runs only when the texture has a `.lex` (true-colour regions are a
+model feature): on lex-less UI sheets it mis-fired on dithered 8bpp art
+(two indices alternating reads as a coherent checker in CT32) and
+repainted the casino slot reels — regression found 2026-09-24.
 Regions failing both readings stay garbled — the remaining static limit.
 
 ## Scene archives `.a` (+ `.fpk`, `.arc`) — ARX-packed members (`browse.py: iter_arx_containers`)
 
@@ -37,6 +37,20 @@ compressed stream, while its pixels do not exist until decompression.
   is the ARX bit stream (the entry-level sweep works because the hair
   words are stream literals).
 
+### Casino regression fixed (raw-CT32 pass gated)
+
+Verifying issue #1 against the July build turned up a regression from
+the 2026-08-21 decoder: the raw-CT32 pass (which draws true-colour
+hair sheets straight from the canvas) also fired on lex-less UI sheets,
+where dithered 8bpp art — two indices alternating — reads as a
+"coherent" checker when taken as CT32. The casino slot reels
+(`tanaka/slot_1`) came out blue/yellow checkered and the poker felt
+(`tanaka/poker_1`) half garbled. The pass now runs only for model-bound
+textures (a `.lex` exists); verified byte-identical output on all 114
+`char/pc` + `char/npc` atlases, clean reels and felt. Still imperfect in
+the casino: `base.png` (slot symbol sheet) is tinted by a wrong overlay
+palette, as it was in every build.
+
 To pick this up: re-run `browse --kinds textures` on an existing dump
 (no re-extract). Pure-Python ARX adds ~2 minutes to the sweep.