Teutonic explores loss-reduction weighted incentive model

Team discussed a revised king-selection algorithm that weights incentives by loss reduction rather than equally. The proposal allocates higher rewards to models with greater loss improvements, reducing variance and discouraging pure copy models that merely increase speed. Both approaches tested were profitable, with loss-weighted distribution showing marginal outperformance.

• •Incentive weights based on loss reduction magnitude, not equal distribution
• •Example: 0.01 loss reduction gets ~32% weight vs. 0.001 gets ~3%
• •Reduces variance and penalizes copy models that only add speed