Revised September 8, 2026. The original article described Classic before atomic staking. Its annualized-yield figure and predictions about the fourth-quarter payout factor have been withdrawn. Classic's payout policy changed at round 1343, so extrapolating the old budget formula no longer describes the current incentive.
What the August comparison measured
The archived figure compares principal reported for legacy weekly rounds with their payout factors through the original August 2026 cutoff. The reported principal approached 811K NMR and the factor fell to roughly 0.089. These are historical round observations; a change in principal does not by itself identify new deposits.
Under the legacy budget formula, adding stake could dilute the payout factor, holding the budget fixed. Comparing factors across periods also requires the payout multipliers, scoring horizon and cadence: 0.089 alone does not determine the amount a model earns or the capital it needs.
Atomic staking changes the incentive
Classic round 1343 introduced individually funded positions. Its configured payout score is 3 × CORR60 + 15 × MMC60, with factor 1 and a return clipped to ±100% of that position's principal. A model's contribution score and its selected position size determine the economic outcome. A falling raw round stake during migration is not evidence of a matching capital withdrawal.
The number of positive-stake model slots fell from 4,261 to 4,098 across those rounds. A 1 NMR cutoff instead reports a decline from 2,015 to 555 because many atomic positions are fractional. Neither count is a unique-account or deposit measure.
What this evidence cannot tell us about returns
The withdrawn yield curve multiplied a pooled weekly position-return statistic into an annual rate. It did not reconstruct each account's funding, overlapping commitments, idle balances, claims or token-price exposure. It therefore cannot support the article's earlier claims about an investor's annual carry or marginal willingness to stake.
Likewise, the proportion of models beating a benchmark is a descriptive comparison on a specified score and sample. A result near 50% does not establish a coin flip, equal strategy quality or the absence of useful contributions. The current Models page compares final MMC60 on matched positive-stake populations and identifies the benchmark explicitly.
For current capital balances see Tokenomics. For the incentive itself, see the payout-factor explanation and the staking-return measurement guide.
Sources: the archived August 20 figure retains its original legacy sample. Cutover counts come from Numerai's public round-details responses captured September 8, 2026. The atomic staking specification describes position funding, score-weighted payouts and settlement.