Confidence is a stake

When you publish, you pick a confidence between 50% and 99%. It looks like a mood setting. It is not. Confidence is where most beginners quietly give points away — because it is a stake, and the system remembers every stake you make.

This page assumes you've met the drift-aware baseline — the skill-free robot your call is measured against. Confidence is the dial that decides how much of that gap you're betting.

What the dial actually does

Confidence scales both sides of your score, symmetrically:

When you're right, higher confidence earns a bigger reward.
When you're wrong, higher confidence takes a bigger penalty.

A correct call at 90% is worth much more than the same call at 60%. And a wrong call at 90% is the single most expensive event on the platform. The dial doesn't just express how you feel — it sets the price of being wrong.

Say two forecasters both call BTC up, and both are right. One published at 60%, the other at 90%. The 90% call earns clearly more skill — it staked more and delivered. But had the asset gone the other way, that same 90% would have been the deeper loss. Confidence cuts both ways, every time.

This is a proper scoring rule — a scoring design with one built-in property: over many calls, the way to maximize your score is to state confidences that match your true beliefs. Round up to sound impressive and the penalties bleed you; hedge everything to 51% and you never earn what your good calls deserve. Honesty isn't a virtue the system asks of you — it's the strategy that literally scores highest.

Why 50% is the floor

You can't set confidence below 50%, and that's not an arbitrary limit. A confidence is a statement about a side. If you thought "down" was more likely than "up," you would simply publish a down call — at more than 50%. So every prediction states the side you think more likely, at how much more likely you think it is. 50% is "barely more than a coin flip"; 99% is "I'd stake my record on this." There is no honest call below the coin flip.

Calibration — the number behind your number

Confidence, aggregated over your whole history, produces something more revealing than any single score: your calibration.

Calibration asks a simple question: of all the calls you tagged around 70%, did about 70% of them come true? A well-calibrated forecaster's confidences mean what they say. If your "90% sure" calls only land 60% of the time, you are overconfident — and the scoring already made you pay for it, one confident miss at a time. If your "70%" calls land 85% of the time, you are underselling good judgment and leaving skill unrecorded.

Your profile shows this as a calibration view — a way to see, across your record, whether your confidence is a promise you keep. It is one of the most honest mirrors on the platform, and it can't be faked: it's assembled from sealed, resolved calls that can never be edited or deleted.

The takeaway

The confidence slider feels like conviction, but it prices like a promise. Set it to what you actually believe. Reserve 90%+ for the calls you'd be genuinely surprised to lose — because on a long enough record, the system will find out whether you meant it.

Next: How a prediction resolves. Or see how confidence and the baseline combine over a career in How reputation works.