feat: net TAO flow for emission allocation

[igoraxz]

Summary

Replace gross user flow ema(buys - sells) with net flow ema(buys - sells) - ema(emission + chain_buys - root_sells) for emission share computation.

Subnets only receive emissions when they generate positive net TAO inflow for the network.

Simulation results (30-day backtest):

• Current gross flow: 55 subnets EMA > 0 → aggregate profit = −362 τ/day (network losing TAO)
• Net flow: 33 subnets EMA > 0 → aggregate profit = +571 τ/day (network gaining TAO)

Rationale

Currently the network subsidizes subnets for gross TAO flow (buys − sells). This heavily misallocates incentives to subnets with negative net TAO flow for the network: e.g. a subnet gets 1 TAO in user buys → receives 4 TAO in protocol subsidy → net TAO flow ("subnet profit") is −3 TAO.

The fix is to change the emission allocator to maximize subnet TAO inflow net of protocol subsidy.

The net protocol cost to a subnet per block is:

net_subsidy = (emission_injected + chain_buys) - root_sells

• Emission injects TAO into pool → network gives
• Chain buys (excess TAO swapped for alpha) inject TAO → network gives
• Root sells (root validators claiming dividends as TAO) remove TAO → network takes back

A subnet that receives 10 TAO emission but returns 3 TAO via root sells has an actual protocol cost of 7 TAO. It should only need to attract 7 TAO in user buys to be net-positive, not 10.

This update also addresses the runaway effect of chain buys — runaway emission cannot be sustained unless a subnet continuously attracts MORE external TAO than the protocol gives it in emission + chain buys.

Flow separation (no cross-contamination)

User flows and protocol flows are tracked in completely independent accumulators:

Flow	Code path	SubnetTaoFlow (user)	SubnetProtocolFlow (protocol)
User buy (stake)	stake_into_subnet → record_tao_inflow	✅ (+)	—
User sell (unstake)	unstake_from_subnet → record_tao_outflow	✅ (−)	—
Emission injection	adjust_protocol_liquidity (direct)	—	✅ (+)
Chain buys (excess)	swap_tao_for_alpha (direct, not via stake)	—	✅ (+)
Root sells	swap_alpha_for_tao (direct, not via unstake)	—	✅ (−)

Root sells are only recorded on the swap path (root validator claims TAO). If a root validator keeps alpha instead of swapping (Keep mode), no TAO leaves the pool and nothing is recorded. The recorded amount is amount_paid_out — the actual TAO that left the pool via AMM, not a theoretical value.

Block-level sequencing

Block N:
  User extrinsics → SubnetTaoFlow accumulates
  Step 4: run_coinbase
    ├─ SubnetRootSellTao::clear()           ← reset audit counter from N-1
    ├─ get_shares_flow()
    │   ├─ get_ema_flow()                   ← fold SubnetTaoFlow (from N-1), reset
    │   └─ get_ema_protocol_flow()          ← fold SubnetProtocolFlow (from N-1), reset
    │   └─ net_ema = user_ema - protocol_ema (if toggle on)
    └─ inject_and_maybe_swap()
        ├─ SubnetProtocolFlow += emission
        └─ SubnetProtocolFlow += excess
  Step 8: run_auto_claim_root_divs
    └─ SubnetProtocolFlow -= root_sell_tao
    └─ SubnetRootSellTao += root_sell_tao

→ Block N's accumulated protocol cost folded into EMA at block N+1
  (same one-block delay as user flows — consistent timing)

Changes (7 files, +127 lines)

New per-block storage (auditable base values):

• SubnetExcessTao — excess TAO swapped (chain buys) per block
• SubnetRootSellTao — TAO from root dividend sells per block

Protocol cost EMA:

• SubnetProtocolFlow — per-block accumulator (emit + chain_buys − root_sells)
• SubnetEmaProtocolFlow — EMA with same smoothing factor as user EMA

Share computation:

• net_ema = ema(user_flow) - ema(protocol_cost) when toggle is on
• Protocol EMA always computed regardless of toggle (stays warm for instant switching)

Sudo toggle:

• NetTaoFlowEnabled (default: on) — sudo_set_net_tao_flow_enabled (call_index 91)
• Set to false to instantly revert to gross flow behavior

Cleanup: all new storage items removed on subnet deregistration.

Test plan

• Unit tests for get_ema_protocol_flow with known inputs
• Verify get_shares_flow produces correct net shares when protocol EMA > 0
• Verify toggle off reverts to gross flow behavior
• Verify new storage items cleaned up on deregistration
• Integration test: simulate emission + chain buys + root sells over multiple blocks, verify protocol EMA converges
• Verify no regression on existing SubnetEmaTaoFlow behavior

🤖 Generated with Claude Code

[JohnReedV]

Would this be more accurate as
Self::record_protocol_inflow(*netuid_i, buy_swap_result_ok.amount_paid_in); instead?

[gztensor]

Should probably use U64F64 here