How it works
Crypto Twitter is a machine for making predictions nobody ever audits. This is the audit layer.
- 1
Commit before the event
Pick a question, state a probability, choose how much conviction to put behind it. It is written on-chain with a timestamp. You cannot edit it, and you cannot delete it.
- 2
Post a refundable bond
A small USDC deposit makes spam expensive. It is returned in full when the question resolves — whether you were right or wrong. Nobody wins money here and nobody loses it.
- 3
It resolves itself
Questions resolve from a Pyth time-weighted average price at expiry. No committee, no arbitration, no appeal. If the oracle fails, the question voids and nobody is scored.
- 4
A record accumulates
Each resolution adds to your Brier score — the squared distance between what you said and what happened. Over months this becomes something that cannot be faked, because the only input is time and being right.
Conviction
You get a fixed budget of conviction points each week. A prediction can be weighted 1×, 3× or 5×, and the weight multiplies its effect on your score in both directions.
This is the sizing decision, the risk of ruin and the public blowup that leverage gives you — with nothing at stake but reputation. A forecaster who never uses 5× is well-calibrated and useless. One who maxes 5× at 95% and is wrong has done something permanent and visible.
What this is not
- Not a prediction market. Your bond comes back whether you are right or wrong. There is no pot, no counterparty, and nothing to win.
- No token, no points, no rewards. A published points formula would fill this with wallets predicting for yield rather than for accuracy, and the dataset is the entire product.
- Nothing for sale. A record is earned and cannot be transferred. You cannot buy a two-year calibration history at any price, because supply is fixed by the clock.
The score
Brier score is the squared distance between what you said and what happened, averaged over your predictions. Say 90% and be wrong and you take 0.81. Say 95% and be right and you take 0.0025. Lower is better.
The raw number means little on its own, so it is shown against a baseline forecaster who always guesses the historical base rate — over the same questions you answered, so answering only the easy ones buys nothing.
No score is displayed until a record has enough resolved predictions behind it. A hot streak is not a record.