Logo UVs used q=p/b, so a non-square half-size stretched the mark
while e∈(0,1). Sample through isotropic p/min(b) (equals q when the
marble is square) and keep ~1px Energise-style SDF AA. No extra
samples or uniforms; final marble look unchanged.
aa=2.5/b ignored d(logoUV)/d(px)=logoScale/b, so the softstep was a
fixed ~10px band — wider than the dots on a ~55px marble and centres
never reached full mask. Use 1.5*logoScale/b like Energise's N/size
AA; keep lower placement and lit^2.4 shadow dissolve.
Nudge the Energise SDF mark down and slightly smaller in q-space so
the three top dots stay on the disk, and fall off logo contribution
with pow(lit) so it dissolves into marble shadow instead of sitting
as a flat grey overlay.
Port the analytical KVT logo SDF from kwin6_effect_kvt_energise
(g22.svg linger mask) into both marble variants. Logo is white and
multiplied by the existing marble lighting so it softens into sphere
shadow; no texture sample and no extra ShaderUniform.
Pack tint id in the integer part of uProgress and morph in the fraction
(with uCrumpleEnd timing) so hue modes keep smooth float32 progress
without a second ShaderUniform. LUT/classHueCache and UI tooltips kept;
default MixColor is Semiwillekeurig for both close and minimize.
Named colours 0–10 unchanged. MixColor 11 = Willekeurig (seq × golden angle → hue),
12 = Semiwillekeurig (FNV-1a hash of window.windowClass → hue). Both use HSV S=1,
V=2/3 (same peak channel as #aa… glass colours), not a fixed extra palette.
Empty windowClass → black. Shader decodes tk>=1000 as HSV. Tooltip explains the
difference. Static Node checks in tools/test-semiwillekeurig-tint.mjs (KWin hand-test still required).