Why we don't publish crypto price predictions

What a hit rate hides

Most crypto price predictions you will find today come with a percentage. Almost none come with the only number that gives that percentage a meaning: what saying nothing would have scored on the same calls.

It matters more than it sounds. Over thirty days, most coins finish roughly where they started — so “no strong view” is simply the most common right answer. An analyst who writes it every single time, reads no news and does no work, still scores. That is the bar. Any real call has to clear it, or it added nothing but confidence.

We published a hit rate for three weeks without publishing that bar. When we finally computed it, our signal turned out to be four times worse than silence.

We checked properly before we decided

One bad quarter is not a verdict — crypto moves in cycles measured in months, and any model can look broken in the wrong season. So we tested the same signal offline against four and a half years of daily prices across 62 coins, roughly forty thousand separate calls per horizon.

It never beat the best constant answer. Not over one day, one week or one month. Not at any threshold. And inverting every call did not beat it either — so it was not a good signal wired backwards. It was noise with a confident arrow on it.

What we publish instead

We publish scenarios, not price targets. For each event that actually matters, we set out what happens if it goes well, what happens if it goes badly, what is most likely, and — this is the part that costs us — what would prove us wrong.

Those calls are graded too, 7 and 30 days later, against the real market move, and the misses stay on the page next to the hits. The score is published beside the same do-nothing baseline you see above, including on the days the comparison does not flatter us.

A scoreboard that can only go up is not a scoreboard. It is an advertisement.