Any claim about an extreme is a claim about a distribution. Funding at some level is only high relative to a history — this coin’s own, on this venue, over some period — and change any of those three and the same number stops being extreme.
Three ways the comparison goes wrong
Wrong asset. Funding distributions differ enormously by coin. A rate that is a once-a-quarter event on a large, heavily arbitraged market can be an ordinary Tuesday on a smaller one.
Wrong venue. Since venues do not share a rate, a percentile has to be computed against the same venue’s own history. Ranking one venue’s print against a blended series compares a number to a distribution it does not belong to.
Wrong window. Rank against thirty days and you find extremes constantly. Rank against two years and almost nothing qualifies. Neither is wrong; publishing the number without the window is.
The sample-size discipline
Percentiles computed off a handful of observations look authoritative and mean very little. Below a floor of observations the honest output is no rank at all rather than a precise-looking figure — the same rule the NoVo track record applies to its own claims, and the reason a bucket that misses its floor renders nothing instead of a number.
This matters more in crypto than elsewhere because history is short. A coin listed months ago simply does not have a distribution yet, and saying so is more useful than manufacturing one.
What an extreme actually justifies
Not a fade — see funding as a crowding gauge. What it justifies is a change in how you size and where you place stops, because an extreme reading means the liquidation structure underneath is unusually one-sided, and a move that starts has more forced flow to feed on.