Suspicious voting patterns relating to account Reputation appear at the very bottom of the sample, and at the top.

You left out one of the most important things to note when talking about Reputation on the steem blockchain.

It's a logarithmic curve. It's a logarithmic curve that can only increase for an individual account by being voted up by a higher Reputation account.

This process is important when talking about anything dealing with Reputation, because it puts an interesting spin on trying to figure out how accounts reached that Reputation in the first place.

Sure, Reputation typically goes along with simply being on the platform longer – but it also simultaneously is only found in an account which regularly gets visibility and exposure to accounts with higher Reputation. Because the curve is logarithmic, accounts with higher Reputation become increasingly rare, at a ridiculous rate, as you go up.

Other analysis has suggested that Reputation 60 is the "magic number" at which the amount earned on the platform rapidly diverges upwards from income consistently garnered before that point. I don't think that is a coincidence; I believe it's correlated with the cohort age and survivability of the wave of users that came in about the same time.

A month over month retention analysis would be difficult purely because of the sheer amount of data to be filtered through, but we know that new user retention hovers between 6% and 8%. The retention rate for accounts which survive up to Reputation 60, much less 70+, have got to be horrendous.

So let's consider what a sock puppet has to do to climb out of the Reputation 25 starter space and move up into the suspected 40 – 44 range from your observations. They have to receive votes from higher Reputation accounts. If we see a Reputation cohort all plateauing around the same level, it's reasonable to assume that if most of the votes involved are going between each other, their visibility is largely only to one another, at least in enough volume to move the needle at that level. It seems to imply that if that plateau is stable, then the sock puppets all have roughly the same Reputation and probably mutually sustain at that level because there aren't enough at a higher Reputation to pull them up further along.

In the cases of specific Reputation ranges that tend to vote for another Reputation level, particularly a higher one, it might be worthwhile to simply extract the whole slice of accounts which have that level of Reputation and examine them individually. At anything above 60, this ought to be relatively straightforward and manageable.

Don't forget, some of this behavior can be explained by there being a significant presence of non-public vote automation going on behind the scenes. It would certainly make sense for someone with a modest amount of server availability and a little bit of liquid steam to make an extended synthetic community of accounts which do very little but vote for one another across the network in order to insert claims into the reward pool on a regular basis. Done consistently, this could be fairly self-sustaining.

I've also come to believe by looking at relationship maps that a number of the accounts with Reputation 70+ are not independent entities, but instead organized and operated by a singular intelligence who has spread to their SP across multiple accounts in order to obscure his or her activity. For that level of Reputation and the amount of SP that we're talking about, that person would have needed to be involved with the blockchain from a fairly early time or have an immense amount of cooperation from one or multiple people who were.

The logarithmic nature of Reputation and the requirement of having someone with a higher Rep vote you up into that space makes certain demands of subsequent analysis.