The exception nobody had chosen
24 October 2026
Every provider here is asked for the same markup: half a percent, one number in one place, and the whole point is that it is the same number everywhere. A board that ranks providers has to treat them identically, or the ranking measures our arrangements rather than their prices.
Eleven of twelve did. One did not, and it had been winning more hops than it should have.
A percentage on somebody else's scale
Most providers hold the markup in a dashboard this code cannot read. You set it once, they apply it, and the quote that comes back already has it in. FixedFloat does it differently: the share is a parameter sent with each request, on a scale of their own devising rather than a percentage of the rate.
So the constant that every other adapter passed straight through had to be converted for this one — and it was not. The parameter carried their own baseline and nothing else. Twelve providers quoted with half a percent taken out; the thirteenth quoted without it.
Why a better price can be the bug
The symptom is the part worth dwelling on, because it does not look like a symptom. One provider sat higher in the rankings than its rate justified. Nobody complains about that. A visitor gets a slightly better number, the provider wins the comparison more often, and every figure on the screen is internally consistent.
It is still wrong, and it is wrong in the direction that matters most here. The comparison exists to say which provider pays best for a given pair. If one of them is quoting on different terms from the rest, the answer to that question is no longer about the providers — it is about which of them we happened to configure correctly.
The fix was arithmetic. The lesson is that a number that favours the reader is not self-evidently right, and an exception that nobody deliberately made is still an exception.
Every figure above can be checked against the providers yourself — no account, no sign-up, and nothing is held for you at any point.