Three times in four

Under the list of providers, in the place where a site explains why it bothers comparing them at all, it said: the best is not always the same name twice.

It reads as a promise of variety. Prices move, providers trade places, and the winner is a different one each time you look — so you had better compare. That is the standard argument for a comparison site, and it is the argument the sentence was making.

Across 1,108 comparisons in which it answered, one provider was the best price 838 times. Seventy-five and a half per cent. Three times in four, the cheapest name is the same name.

The denominators differ on purpose and are worth reading carefully: each is the number of comparisons that provider actually answered, not the number that took place. A provider that replies rarely can hold a respectable share of the contests it entered while winning almost nothing overall. Comparing the percentages across rows would be the second mistake in an article about the first.

Nothing in that sentence is false. The best is not always the same name — one time in four it is somebody else. A lawyer would pass it.

But it was doing work the data does not support. A reader takes “not always the same name twice” to mean the leaderboard shuffles, and it does not shuffle. It has a clear favourite and a long tail, and the honest description of that is not the one that was printed.

This is the exact failure this site is willing to be tiresome about when other services commit it. One competitor advertises privacy routing and declines to say which chain the money crossed. Another says “no KYC, no account, full self-custody” while executing on venues that require all three of whoever they deal with. Both statements are defensible word by word. Both leave a reader believing something untrue. Ours was smaller and the same shape.

Not by an audit. The figures were being read for something else entirely — a question about whether a new provider’s default commission matched what the others charge — and the winner’s share was simply sitting there on the way past.

That is the uncomfortable part. The dataset that contradicts the sentence is published by this site, on a page linked from the sentence itself, and had been for weeks. It was one click away from the claim the entire time, and nobody had made the click.

The correction is not an apology, because the accurate version is a stronger case than the one it replaces.

“The leaderboard shuffles” is a weak reason to compare prices. If the winner were genuinely random you would be as well picking one and accepting the average.

What is true is better: one provider usually wins, a quarter of the time it does not, and nothing tells you in advance which quarter your transfer falls into. The only way to know is to ask all of them — which is not a slogan about variety, it is a description of the product. The front page now says that instead.

No figures were put in the new sentence. Numbers written into prose go stale quietly, and the ones above are republished from the measured record every fifteen minutes. If the favourite changes, that page will say so before this one does — and this article will be the thing that is out of date, which is the right way round.