Flavius Josephus's sieve: Start with the natural numbers; at the k-th sieving step, remove every (k+1)-st term of the sequence remaining after the (k-1)-st sieving step; iterate

Open in the 3-D viewerA000960 on the OEIS
| Terms | 100,000 (n = 1 … 100,000) |
|---|---|
| Decomposable (a > 2d) | 99,997 |
| Level class, k > L | 96,561 · 96.56 % |
| Weight class, k ≤ L | 3,436 · 3.44 % |
| Ties, k = L | 2 |
| On the level line L = 1 | 13,258 |
| Forced level, l ≤ d² | 77,425 |
| Range of a(n) | 1 … 7,854,038,953 |
| Range of the jump d | 2 … 521,092 |
| Largest weight k, level L | 7,853,264,669, 5,149,949 |
count(x) = 2 sqrt(x/pi), so a(n) ~ pi n^2/4 and the gap outgrows sqrt(l): l <= d^2 on 77 % of terms, and 96.6 % of the sequence is level-classified. It is the spread of the gap, not its mean, that populates the weight class.
The decomposition
Every term of a strictly increasing sequence is written a(n) = k(n)·L(n) + d(n): the jump d = a(n+1) − a(n), the weight k the least divisor of a − d greater than d, the level L = (a − d)/k. A term decomposes when a > 2d; it is in the level class when k > L and in the weight class when k ≤ L. See decompwlj.com and arXiv:0711.0865, or how it works, with worked examples and a live one.