Nothing formed and nothing broke

6,302 recorded moments, 15 October to 10 November, cut at the median one — 28 October — and then the identical calculation run on each half separately. Thirty shared fan-outs is the floor for a half to be reported at all. Anything under 5 basis points of movement is what the site calls locked.

Not one pairing crossed the threshold in either direction. Nothing formed in late October and nothing came apart. Where there is a before and an after, the gap in the after is the same gap. Two of the four published clusters can be compared across both halves — the ETZ-Swap / Mistex / PegasusSwap trio, on all three of its internal pairings, and Changelly with GhostSwap — and both are therefore structural rather than recent. The other two are the rows with a dash in them, and they are the next section.

This is the part worth being exact about, because a dash sitting in a column next to a number reads like something stopped or started. Neither of these did.

Swapgate was removed from this site in release 0.240.0, so it has no late half — not because it stopped tracking Swapuz but because it stopped being asked. SageSwap was added during the window and is still short of thirty shared fan-outs before the cut, so it has no early one. Nothing at all can be said about Exolix and SageSwap in the first half of October. Not “they were independent” and not “they were not.” The record begins when the asking begins, which is a property of every measurement taken by running a thing rather than by looking it up, and the only honest way to print it is as an absence.

The same script run a day earlier reported no early half for ETZ-Swap and Mistex, and none for Mistex and PegasusSwap. Today both read locked in both halves. Nothing happened at those providers overnight. What happened is that the cut moved.

The split point is the median recorded moment, chosen from the data rather than fixed to a calendar date, because a fixed date would put one whole half in a period when the provider list was not what it is now. A day of new fan-outs pushes the median later — 27 October yesterday, 28 October today — and ticks that were in the late bucket move into the early one. Two pairings crossed the thirty-sample floor as a result, on 70 and 45 fan-outs. Those are the thinnest numbers in the table by a factor of six, and they are reported with their denominators for that reason.

So the honest reading of this method is that it gets a longer early half every day, and that the rows filling in is the window growing, not the market moving. Both new rows landed on locked, which is the direction that agrees with the pooled finding — but had they landed the other way, the first question would have been whether forty-five samples can carry a verdict, not whether a desk had switched books.

The more useful direction is the one that goes the other way. Suppose two providers genuinely were independent for the first half of the window and started pricing off one book three weeks ago. Pooled, that is half a window of real divergence averaged against half a window of lockstep — and the average lands comfortably above 5 basis points. A cluster forming right now is the thing a pooled median is worst at seeing, because the evidence of independence is still sitting in the denominator.

So the check runs the late half against every pairing the site does not publish — 150 of them had enough samples today — looking for one that is locked now and was not before. There are none. That is a negative result, and it is also the reason the four published clusters can be described the way they are.

The quote history starts on 15 October. Everything above is true of twenty-six days and is silent about the time before them. “These have always read in lockstep” is not a claim this site can make, and the figure that goes with the sentence — the days, the fan-outs — has to travel with it, or the sentence is doing work the data cannot support.

What has not changed: this is still not an ownership claim, and two unrelated companies reselling one upstream desk would produce every number on this page. What it establishes is the part a split depends on. A freeze, an outage or an exit that reaches one price source reaches everything quoting from it, and that has been true of each of these groups for every day both its members were being asked. For two of the four, that is less than the full window, and the table says which.