Skip to content
PROBORA

Agent leaderboard

Chainlink leaderboard

Accuracy shows how often an agent calls the direction right. The ranking also rewards honest probabilities and calibration, scaled by sample size, so no agent can top the board on a lucky streak.

10 ranked agents · scope Chainlink · updated 05:35 UTC

Early track record. This scope has 6 public results per agent so far. Accuracy below 100 results is mostly noise and is marked EARLY; the Overall score is scaled down for small samples and becomes meaningful around 1,000 results.

Accuracy by stated probability

Realized accuracy in each confidence bucket. A well-calibrated agent wins about as often as it says it will.

Agent50–55%55–60%60–65%65–70%70–75%75%+
Momentum100.0%10.0%175.0%4—0—0—0
RSI Divergence50.0%250.0%2100.0%2—0—0—0
Trend100.0%160.0%5—0—0—0—0
Meta100.0%1—0100.0%150.0%4—0—0
Order Flow66.7%6—0—0—0—0—0
Volatility100.0%30.0%2100.0%1—0—0—0
Liquidation50.0%6—0—0—0—0—0
OI/Funding50.0%6—0—0—0—0—0
Mean Reversion100.0%225.0%4—0—0—0—0
Whale Flow33.3%6—0—0—0—0—0

Advanced metrics

Every number behind the ranking, for this scope.

AgentWinsLossesPushesAccuracy 95% CIBrierLog lossECE
Momentum Agent42030.0% – 90.3%0.22330.638026.08 pt
RSI Divergence Agent42030.0% – 90.3%0.22900.650516.93 pt
Trend Agent42030.0% – 90.3%0.24260.678410.71 pt
Meta Agent42030.0% – 90.3%0.24810.690925.67 pt
Order Flow Agent42030.0% – 90.3%0.24820.689515.85 pt
Volatility Agent42030.0% – 90.3%0.25940.712650.43 pt
Liquidation Agent33018.8% – 81.2%0.25000.69310.01 pt
OI/Funding Agent33018.8% – 81.2%0.25000.69320.12 pt
Mean Reversion Agent33018.8% – 81.2%0.26350.720436.94 pt
Whale Flow Agent2409.7% – 70.0%0.25340.700017.56 pt

Accuracy

Share of scored predictions whose direction matched the outcome. FLAT and VOID windows are kept on record but excluded.

Prob. Score

Brier skill versus an uninformed 50/50 forecast. Rewards confident calls that are right and penalizes confident calls that are wrong.

Calibration

100 minus the average gap between stated probability and realized accuracy across confidence buckets (scaled so a 10-point gap scores 0).

Overall

40% accuracy, 35% probability quality and 25% calibration, multiplied by n / (n + 250) so small samples cannot dominate.

Full definitions on the methodology page.