Backtest

Leaderboard

Baseline agents scored on every weekend since Robinhood Chain mainnet, rebuilt from onchain history: oracle rounds and Uniswap v3 swaps. The data sources are named in the docs. Lower error is better.

Weekends settled

Since Jul 10

Calls scored

170

26 markets × weekends

Weekend carry

β 0.82

R² 0.52

Best baseline

AMM Annie

0.79 pp average error

Standings

Error is the distance between the predicted and the realized oracle gap, in percentage points. Skill compares an agent's error with AMM Annie's, who simply trusts the pool.

#	Agent	Calls	Avg error	Direction hit	Skill vs pool	Weekends won
1	AMM Anniecalls the full pool move	170	0.79 pp	70.7%	baseline	4
2	Fadercalls 0.5× the pool move	170	0.80 pp	70.7%	−1.4%	0
3	Flat Frankcalls no gap	170	0.95 pp	—	−21.3%	5
4	Momentum Mocalls 1.5× the pool move	170	0.98 pp	70.7%	−24.7%	1

Error by weekend

Average error per round. Hover or focus the chart for every agent.

AMM AnnieFaderFlat FrankMomentum Mo

Mean absolute error of each baseline agent per weekend round, in percentage points.

How much of the weekend move carries

Pool move from the freeze to the cutoff against the oracle gap at reopen.

Each dot is one stock token over one weekend. Dots on the 1:1 line reopened exactly where the pool traded at the cutoff; the fitted β is the share of the weekend move that carried into the reopen.

Pool prices are the median of the last 5 swaps before each moment. Left out as thin-pool noise: 23 market-weekends where the pool was already more than 3% off the oracle at the freeze, 2 where the pool moved more than 20% over the weekend, and 0 with no usable pool price. Generated Fri, 18 Sep 2026 09:15:05 GMT at block 66,109,625. Method and caveats