A 25-delta option is loosely described as having about a 25% chance of expiring in the money. The general relationship, and its caveats, are in delta and probability.
It is a convenient approximation of a model output, not a measured frequency — and the model assumes returns are distributed in a way crypto respects less than most markets.
The fat tails problem
The pricing model assumes a distribution with thinner tails than crypto actually has. Large moves happen more often than the model expects, partly because of the leverage feedback in why crypto volatility clusters.
So far out-of-the-money options are more likely to finish in the money than their delta suggests. The market knows this and prices it, which is why the volatility surface is not flat — and it means delta-as-probability is least reliable exactly at the strikes where people most want a probability.
Settlement changes what delta even means
On a coin-settled contract, delta is expressed against the coin while the payoff is curved in dollars. So a 25-delta call is not a 25% chance of a proportional dollar gain — the relationship is bent by the inverse payoff.
Comparing a 25-delta strike on a coin-settled book with one on a linear book is therefore comparing two different things, the aggregation trap from settlement currency decides what the numbers mean.
What it is still good for
Comparing strikes on the same book, which is the job it actually does well. Delta-based strike selection is a consistent ruler even when its absolute interpretation is loose — and skew is defined as a comparison at equal deltas for exactly that reason.
The framing
Treat delta as a position on a ruler, not as a forecast. Every use that survives contact with crypto is a relative one.