A model can move from a large legacy NMR tier to a small atomic tier without withdrawing capital. Across the Classic cutover, the principal assigned to a round changed dramatically while positive-stake model counts changed far less.
Revised September 8, 2026. Earlier pooled tier rankings are withdrawn as a cross-regime comparison; they did not hold the capital basis constant.
Define the question before the tier
If the question concerns influence within one round, rank models by their share of that round’s scored principal. If it concerns capital commitment, use measured balances or clearly labeled allocation estimates. An absolute position tier cannot answer both questions across the migration.
For outcome comparisons, use only settled, positive-stake, non-benchmark observations. Net NMR divided by the principal of those same observations is a position-return measure. Model quantiles and the stake-weighted pool ratio answer different questions; neither is automatically an account return.
Avoid retrospective selection
Selecting today’s largest models and plotting their history describes a current cohort’s past. It does not describe who held the largest shares in each earlier round. The Models page identifies retrospective cohorts and keeps return windows within a staking regime.
An apparent tier premium can also reflect different score exposures, participation histories or survival. I would require stable definitions and cohort sensitivity before claiming larger stakes identify more or less skill. The score comparison guide explains why raw score coefficients are not enough.
Method: the replacement figure uses Classic round 1342/1343 model lists captured September 8, 2026. It demonstrates tier migration, not an estimated performance premium. See round-return denominators.