I’m missing something here… what do you mean by “advance both sides?”
> an informative period failure is 15 = 6 + 6 + 3. Doesn't this conflict with something in the article?
You’re talking about, say, P(x) = x^4 + x^2 + 1, for a four-bit LFSR? This will have three nontrivial cycles, with length 6, 6, and 3. Rather than talk about this specific polynomial / specific set of taps, we can consider that the initial (non-zero) state is in one of those three cycles. There are three possibilities for the length of a cycle:
1. Maximum period 2^k-1. In this case, 15.
2. Period which is a factor of 2^k-1, such as 3.
3. Period which is not a factor of 2^k-1, such as 6.
We can distinguish these case by picking an initial state and then doing a series of checks, where we advance the LFSR by N steps, and then see if it is in the initial state again.
For exactly the LFSRs with maximum period, the LFSR will be in the initial state after advancing 2^k-1 steps, but it will not be in the initial state after advancing (2^k-1)/p steps for prime factors p of 2^k-1. In the case of a 4-bit LFSR, we check N=15, N=5, and N=3. The cycles of length 6 will fail the test because they will be in a different state after advancing 15 steps. The cycles of length 3 will fail because they will be in the same state after advancing 3 steps. The end result is that the LFSR with cycles of length 6, 6, and 3 will be rejected, no matter which cycle the initial state is in.
> tradition makes people produce LFSRs with very few taps but if you're doing it in software there's no real reason you have to do anything other than keep them even and use the end bits.
The reason I was looking at this in the first place was out of a kind of retrocomputing interest, and using something like 0x100001b as your polynomial for a 24-bit LFSR can make the implementation slightly shorter / faster, on some older CPUs and with memory constraints.
It was easy to find a lot of handwaving and code / numbers copied from place to place, and I wanted to know how to generate the polynomials / pick the taps in the first place. The program I wrote just goes through the polynomials in lexicographic order, so it’s not actually trying to pick the fewest number of taps, but it will prefer sets of taps where the taps are all clustered on one side.