The more a transaction offers, the higher it sits and the brighter it burns. The axis is logarithmic and rescales with the market.
The floor bandat base fee
Below the lowest rule is the band: transactions offering nothing above the base fee. In every mainnet recording so far that is most of the pool. A mark's height inside it means nothing — how full it is does.
Fadingtime waiting
A mark dims the longer it waits, whatever it pays. Brightness is not age and age is not brightness: the two never share a channel.
Row widthgas used, on a root scale
In the block, each transaction is a row as wide as the gas it consumed, against the largest in that block. Square root, so a rollup batch does not flatten every transfer onto the floor.
Tone and trackseen, or never seen
A line reaching a row is where that transaction came from in the mempool. A warm row has no line: this feed never saw it pending. That is the only thing the warm colour ever means here.
What this measures
An Ethereum transaction is normally announced before it is settled. It is broadcast to the public mempool, where it waits among tens of thousands of others while builders decide what to include. Anyone running a node can watch that queue. It is the left of this screen: every mark is one transaction still waiting, placed by the fee it is offering.
Every twelve seconds a block is built, and some of what appears in it never went through that queue — not as this feed saw it. It went straight to a builder, through a private relay or inside a bundle, and was public for the first time when the block was. Those transactions cannot fly in from the left, because they were never anywhere: they appear in place, and they are the only warm colour on the screen.
The share matters because the mempool is the part of Ethereum that is observable in advance. A transaction that skips it cannot be front-run, which is often exactly why it skipped — and it also cannot be seen, priced, or argued with until it is already final. The figure above is how much of one block this feed never saw coming.
What it cannot tell apart
“Never seen” is a statement about one vantage point, not about the network. Four things land in the same count as private flow, and the instrument cannot separate them:
Propagation. A single node sees most of the public mempool, not all of it. A transaction that reached other nodes but not this one before it landed is counted here as never seen. This is the largest source of error, and it is not measured yet.
Memory. A hash is remembered for five minutes. A transaction that waited longer than that and then landed is counted as never seen, although it was public the whole time.
The cold start. Nothing is classified until the pool has been watched for five blocks, or seeded from a snapshot of it. Until then the figure is blank — not zero, because zero would be a claim.
Reorgs. When the chain replaces a block, the replacement is classified on its own and the orphaned block is withdrawn from the record, so a transaction that lands twice is not counted twice.
So the figure is an upper bound on private flow as seen from here, with the first term unquantified. Measured against an independent record — a second node, or a public mempool archive — it would become an estimate. It has not been.