The first graph , from the paper, is not the number of transactions. This is the number of edges.
I dont know what that is and I have never heard that term before in the blockchain.
It's written in the paper that in their Transaction Graph they represent addresses (u may consider them TXOs old & new) as vertices and TXs as edges connecting these addresses(from input to output), sorry I had to add this quote from the paper in my post to make it clearer.
In the transaction graph of a cryptocurrency, vertices are accounts (or addresses) in the currency network, and the edges between them are transactions between those accounts.
So it's supposed that the cumulative no of edges is the same as the cumulative no of Transactions, unless they count some more than once or in a wrong way.
.
In the image you posted there is no mention to "miner merge" . The paper you linked before is merging transactions and addresses from the same individuals , is that what you call miner merge?
.
I'm sorry the term "miner merge" also appears inside the paper for case I in the Fig (saying it could be used to identify that the resulting UTXO is a miner address (like cold wallet address that contains a lot of money)
From the paper:
From mid 2010 to mid 2011, the Bitcoin exchange rate starts
to climb, more and more miners begin to merge Bitcoin (Case
I in Figure 10) in order to sell it for profits......
.....
In Case I, where the Bitcoin is
untouched since 2010, the miner accumulates its three mined
Bitcoin addresses into one address. Back in 2010 the rate of
Bitcoins is low, this behavior is perceived as merging Bitcoin
for ease of management
.
Why are you looking in testnet data, not the bitcoin network data?
Because it would be much harder to test a Merkle Tree design on larger datasets, even the MIT Utreexo project I think uses test net data (from the no of UTXOs in the run). Still, I need to be sure it reflects the same frequency of different kinds of TXs to give the same av life span of a single UTXO.
.
On 2nd thoughts, maybe the test net is actually very accurate even in representing time-behavioral changes; maybe Block381 is supposed to refer to 2010-2011 and that's what miners used to do then.