Pretty Pictures Tell a Thousand Tales
Yes, yes – I know. I'm supposed to be writing more about role-playing games and other aspects of ludology.
But here's the thing. Once upon a time, many moons ago, I used to be a professional programmer. People actually paid me to look at bits of code, figure out where they were broken, and come up with solutions – not necessarily ones that required modifying source code.
Crazy, I know.
Unfortunately for me, that was representative of a deeper appreciation of enjoying the chase, hunting the elusive bit of information, finding new ways to present it. I have acquired a certain set of tools in my life; tools that make me a nightmare for people like you.
You know, people who would rather read things that aren't in-depth analyses of the processes behind the steem blockchain.
Nevertheless, I persevere!
It occurred to me as I was tinkering about with yesterday's imagery that I had never seen anyone do a serious, data-heavy study of the relationships between the top end of steem blockchain accounts. This might be because the tools are a little abstruse and not exactly easy to wrap your head around. It might be because to do a real, serious study requires some serious computing platform power in order to actually look at the data. It might be because it's boring as crap.
Regardless, that's what I've been up to.
First Cut: No Removals
There were some very, very pretty pictures generated as a result. Not surprisingly, it revealed some definite loci of activity just talking about transactions.
As nice as that was, I thought I could do better.
Firstly, let's expand our scope out to the top 200 accounts by SP. Why not? We've got plenty of horsepower sitting here next to my knee and it's not like the data is hard to cull.
Of course, if we're going to pull the activity of the top 200 accounts by SP, we're going to be looking at transactions. If are going to be looking at transactions, we are also going to need to be looking at all of the accounts that have engaged in transactions with those top 200, which widens our scope just a little bit.
While we're at it, let's color each of those transactions depending on which direction they're going. If they're going between any of the top 200 accounts, let's make that transaction black. If they're going between accounts in the top 200 and accounts out of the top 200, that is value which is moving away from the core, let's make that transaction green. If the transaction is going from one of those ancillary accounts into the top 200, let's make that transaction red.
I admit to this not being a completely arbitrary decision. I want to look at the flow of value in and out of the people that hold the most of it on this blockchain.
So let's look at that naïve solution.
(Click image to go to scrollable, zoomable gigaimage.)
Well. That's enormous.
It also has certain problems. I mean, those might be representative of certain problems in transactions on the blockchain, but they're problems nonetheless.
Immediately, we can see two enormous nexi of activity, one in the south (
There are a few other very obvious features which draw the eye. 
But this is kind of a mess. Can we do better?
Second Cut: A Good Pruning
After studying the previous chart for a fairly extended amount of time, I had to give serious consideration to cutting out a handful of accounts which we had previously considered as "of interest."
All four of them or simply overwhelming my ability to make readable graphs out of the content in the database. One of them is the biggest exchange on the blockchain. Two of them are essentially promotion systems. And the last of them is Jerry Banfield, who is a little bit of all of the above.
This is really where we made the jump from studying about 100 accounts to the full 200 top SP-rated accounts because I thought that if I was going to be trimming some of the big users anyway, I might as well add some more. Considering that cutting just those four accounts from our accounts of interest reduced the number of transactions that we were studying by a literal order of magnitude (that's a factor of 10 for you nonscientific folks), it seemed like a reasonable trade-off.
But that brought us to a little bit of a different place.
(Click image to go to the zoomable, panable, gigapixel version.)
This is still a really busy place, but we again see certain patterns of behavior and islands of isolation.
In particular,
It's not surprising to see 
In fact, the relative absence of transactional energy flowing out of the central network gave me a certain amount of pause. There are an absolute shit ton of transfers going on here, but we don't really have enough context to make good decisions about what we are seeing.
Which leads to yet another round of deep cuts in the data supply.
Third Cut: Charming?
Alright. We have the top 200 accounts by SP marked as of interest, minus a few whose transactional load is largely unrelated to activity that we're interested in. We know what they are.
Most of those transactions are transfers pushing some sort of fund, whether it be STEEM or SBD up into those rarefied reaches.
What if we make two further cuts?
Firstly, let's ignore any transaction of less than 0.5 of whatever is being transferred. That will drop out all of those 0.001 SBD memo spams that I understand whales receive a ton of and help clear out our research space.
Secondly, let's filter out any ancillary account valued at less than 100 vests. That won't actually cut the data feed very much, but it should help cull some of the bots that do nothing but send messages all day with delegated funds.
What's that look like?
(Click on the image to go to the somewhat smaller than the others gigapixel version.)
Right off the bat, there are some obvious differences.
This image is a lot smaller, for example. It's only 23 megapixels while the other two were pushing 150 megapixels each. This is a much tighter, more constrained space.
We can actually see that some of the whales haven't actually engaged in a qualifying transaction of transferred funds in the last month. Or at least nothing that qualifies to trip our particular breaker for noticing. This is awesome. We're getting somewhere.
You can actually make out some isolated transactional spaces. 
Again,
I find of the central position of 
Locality on these graphs represents centrality in terms of distribution of transactions. Nodes that are close to one another tend to transact with nodes that are close to them.
Now having corrected for magnitude of transfer transaction, what do you see?
Even correcting for that and allowing for that, most of the transactions involving whales on the blockchain are red. That is, they represent people who are not whales transferring some amount of currency greater than 0.5 units to the whale.
At least as far as literal fund transfers, we can tell that the majority of the power players, the top 200, sorted by SP, are definitely not dispensing funds to other people on the blockchain. Funds are moving up to them, not down away from them.
That's definitely interesting.
Epilogue
Look, I know I'm no kind of #BusinessIntelligence guy. I leave the serious job of making sense of financial aspects of blockchain operations to other people who are more inclined to be awesome than I am.
Know your strengths.
But I think that I have a few tools to bring to bear on the question of how transactions are flowing across the public blockchain. That's probably a curiosity that's worthwhile to pursue.
The obvious next step is to do some sort of study of the way votes are traveling on the blockchain. That's going to be a much larger and more intractable problem because of the sheer number of votes which happen per minute.
We may can cut down on the amount that we abuse the backend database by deploying some kind of clever query. It should be possible to push the filtering work off onto the server by sending it a list of the accounts we are interested in in terms of vote transactions, but that will require some more research on my part.
If this sort of thing is interesting to you or if you learn something, let me know. We will probably all look askance at you as a result, but it's the sort of thing that keeps me face down in database coding for days at a time, and I think we can all agree that keeping me occupied and off the streets is the best for society in general.
Tools
- SteemData
- graphviz
- Python
(Don't look like that; you're lucky it wasn't written in Erlang) - Jupyter Notebooks
(Seriously, a Hell-send. How did I ever code without this?) - AMD Ryzen 7 1700