YOUR MODEL MARKET
One prompt. Competing models. Your pick teaches the router.
Compare anonymous answers, then choose the one you would use.
Four automated desks lock probability forecasts and stake 100 simulated credits each before inference starts. Other signed-in participants can bet 1–100 play credits during the 10-second window on prompts explicitly shared by their owners. Owners cannot bet on their own markets.
Winning tickets split the pool; a 2% fee applies to winning profit. Rejected answers and pools with no winning ticket refund stakes. Brier score measures probability accuracy; lower is better. Desk strategies are simple priors in v0, not independent human experts.
Each credit settlement includes its sequence, the previous receipt hash, and a SHA-256 hash of its contents. This helps detect edits when receipts are retained independently. The server owns the ledger; this is not a blockchain or a decentralized guarantee.
Measured response times come from the provider. Displayed API costs come from provider results or logs rounded to $0.0001; missing charges stay unknown. Play-credit fees do not pay the inference bill.
One server locks bets, records your verdict, pays winners, and keeps a fee receipt.
A contract holds collateral and enforces payouts. Signed forecasts and model calls stay off-chain; outcome receipts settle in batches.
A signature proves who selected the winner, not that their preference was honest. Independent evaluation and anti-collusion rules would be necessary before real-money bets. No wallet connection, cash bets, token, or chain deployment is active here.