A positive contribution score does not by itself establish a positive payout. For Classic atomic round 1343, the weighted combination is 3 × CORR60 + 15 × MMC60. A negative value in one component can outweigh a positive value in the other.
Revised September 8, 2026. The former CORJ60-based universal formula and its decile-based payout interpretation have been withdrawn.
Contribution is evaluated against a reference
CORR60 evaluates predictions against the 60-day target. MMC60 evaluates their residual contribution relative to the meta-model. They answer different questions, but neither is an observed hedge fund return. The reference model, target, universe and score version are part of the definition. Numerai’s MMC description explains the scoring construction.
Holding CORR60 at +0.02 gives a contribution of +0.06 before clipping. MMC60 at −0.004 offsets that exactly: 15 × −0.004 = −0.06. This is a break-even example under the stated policy, not an estimate of the frequency of profitable models.
Names that look similar can encode different targets
The audit found historical rounds where mmc and mmc60 differed for every observed pair. A chart cannot relabel the older column as MMC60 because recent compatibility fields happen to match. The Trends page keeps explicit 60-day fields and historical diagnostic series separate.
A coefficient of 15 also does not establish that MMC60 explains five times as much realized payout variation as CORR60. That requires the distributions, covariance, clipping and matched principal of the scored sample. The break-even note separates this arithmetic from a capital-return claim.
Method: the payoff curve is an illustration from round 1343’s published configuration, with factor 1 and a ±100% position-return clip. Both input scores are hypothetical; no model or account return is estimated.