Answering is not the same as being worth asking
7 November 2026 · figures measured 7 November 2026
Every time somebody prices a swap here, every provider is asked the same question at the same moment and all of the answers are kept — including the losing ones, which is the only part that is unusual. There are 5,077 of those fan-outs now, across eighteen providers. They are published free, without a key, at /api/measured, and the dataset page says what each field measures and how to cite it.
The dataset answers a question the industry does not: not which provider is cheapest today, but how far behind the best quote each provider sits when it does answer. Five thousand fan-outs in, the answer is strange enough to be worth a post. The two things a buyer would most want ranked together — how often a provider answers, and how good its answer is — turn out to be almost unrelated. And only one of them is visible on any comparison site, including, by default, this one.
The two rankings do not agree
Answer rate is the share of fan-outs a provider replied to. Shortfall is its mean gap to the best quote within the same fan-out — the same pair, the same amount, the same minute, the same set of rivals. Best is how often it won one.
| Provider | Answers | Mean behind best | Was best |
|---|---|---|---|
| FixedFloat | 99.9% | 1.27% | 34.9% |
| ChangeNOW | 97.3% | 9.41% | 1.5% |
| StealthEX | 83.2% | 2.94% | 0.1% |
| Changelly | 72.5% | 0.21% | 35.0% |
| GhostSwap | 70.5% | 1.02% | 0% |
| swapgate | 59.6% | 0.34% | 35.7% |
| Exolix | 46.4% | 0.47% | 2.8% |
Read the first and fourth rows together. FixedFloat answered 5,070 of 5,077 fan-outs and was 1.27% behind the best when it did. Changelly answered 3,674 of 5,070 — roughly three times in four — and was 0.21% behind. They won almost exactly the same share of the fan-outs they entered: 34.9% and 35.0%.
So one of them is available six times out of six and six times further from the best price, and ranking these providers by the number a visitor can actually observe puts the better-priced one fourth.
The second row is the one to sit with. ChangeNOW answers 97.3% of the time, which makes it look like the dependable choice, and it sits 9.41% behind the best quote in the same fan-out. It is the most consistently available provider on the board and the most expensive by a factor of seven over the next-worst.
Never once the best
Two rows in that table are not approximations.
GhostSwap 3,572 answers best 0 times (0%) StealthEX 4,216 answers best 3 times (0.1%)GhostSwap has answered three and a half thousand fan-outs and has never been the best price in any of them. It is not a bad provider by the usual reading — 1.02% behind on average is mid-table, closer than FixedFloat — but there has not been a single moment in 5,077 recorded comparisons where it was the one to take.
That is a fact no amount of rate-shopping at one provider could produce, and it is not observable from any single quote. It only exists because the losing answers were kept.
The figure that moved without the provider moving
Three weeks ago the same dataset said FixedFloat was best 79.2% of the time and 0.42% behind. Today it says 34.9% and 1.27%. FixedFloat did not get worse at anything.
Seven providers were added in between. A provider’s win share falls arithmetically when the field grows, and its shortfall rises for the same reason: the thing it is measured against improved. Both numbers describe the fan-out, not the provider alone.
This is the one way this dataset can mislead, and it is a property of the method rather than a defect in it. A shortfall figure is comparable only against the providers asked alongside it. The alternative — a margin against a market mid-price — looks more portable and is worse, because it silently compares providers across different pairs, different amounts and different minutes. The providers page explains why it does not print one.
What the denominators are for
Every rate above is published with the count it came from, and the counts are not the same, because providers were integrated on different dates:
FixedFloat 5,077 fan-outs 57 pairs swapgate 1,810 fan-outs 44 pairs Mistex 2,213 fan-outs 26 pairs CCECash 1,200 fan-outs 7 pairsCCECash has been the best price 42.3% of the time, the highest share on the board, and it has done so across seven pairs. That is a real figure about a narrow range, and printing it beside a provider measured over fifty-seven pairs without the denominator would turn it into a claim it cannot support. Providers below thirty fan-outs are absent from the dataset entirely rather than reported with a rate.
A number with no denominator behind it is a marketing figure. That is most of what this industry publishes about itself, and it is the reason the counts are in the API response rather than in a footnote.
Why it is free
A dataset like this is the obvious thing to sell, and selling it is structurally impossible here. A buyer needs an account, an invoice and somewhere to send a key, and this site does not have accounts — not as a feature it has deferred, but because the whole product rests on not having them. So the choice was publish it or sit on it.
It is published under a licence that asks for nothing: free to use and cite, attribution appreciated and not required, with the method stated in the response body so a citation can carry its own caveats. Every fan-out this site has ever recorded is in it, including the ones where the provider this site earns most from was beaten.
If you want to check it
curl https://www.crypto-train.com/api/measured returns the whole thing with no key, no rate limit and a Dataset declaration. The rendered version, with each provider’s KYC grade from kycnot.me beside its price record, is at /providers.
Figures in this post were read on 7 November 2026 and will have moved by the time you read them. That is the point of publishing the endpoint rather than only the numbers.
Every figure above can be checked against the providers yourself — no account, no sign-up, and nothing is held for you at any point.