feat: normalize protocol cost in net flow EMA

[igoraxz]

Summary

• Normalizes protocol EMA so its positive total matches user EMA positive total before computing net flow, preventing subsidy concentration
• Fixes NetTaoFlowEnabled toggle injects accumulated SubnetProtocolFlow into the EMA as a single block #2667: protocol EMA now updated unconditionally (accumulator no longer shocks on toggle)

Formula

norm_factor = min(1, user_positive_ema_sum / protocol_positive_ema_sum)
net_flow = user_ema - norm_factor * protocol_ema

• norm_factor self-adjusts each block: more concentration -> it drops -> softer deduction -> rebalances
• Only positive protocol cost is scaled by norm_factor; negative protocol cost (root drain > emissions) passes through at full value
• When NetTaoFlowEnabled = false, norm_factor = 0 (pure gross flow)

Guarantee

Non-negative aggregate net signal over gross-positive subnets. Individual subnets can still be filtered when their protocol cost exceeds their user demand -- the guarantee is on the set, not per-subnet.

Proof: sum of net_i over {user > 0} = sum of user_i - norm_factor * sum of protocol_i. Since sum of protocol_i (over user > 0) <= protocol_positive_ema_sum, and norm_factor * protocol_positive_ema_sum <= user_positive_ema_sum (by construction), the sum is non-negative.

Simulation

41-day simulation with full emission redistribution:

• Gross flow: 70 -> 53 subnets
• Net flow (K=1, no normalization): 70 -> 10 subnets (concentration collapse)
• Net flow (normalized): 70 -> 46 subnets (stable, self-adjusting)

Test plan

• Verify protocol EMA updates every block regardless of NetTaoFlowEnabled
• Verify norm_factor = 0 when NetTaoFlowEnabled = false (pure gross flow)
• Verify norm_factor in [0, 1] when enabled
• Verify steady-state: with constant flows, net approximates user - protocol (norm_factor -> 1)
• Verify concentration resistance: fewer qualifying subnets -> norm_factor drops

Generated with Claude Code