Emission weights

Every subnet carries a five-slot emission weight vector

[burn, recycle, owner, validator, miner]

that says how each block's alpha emission is divided. Each slot is a u64 share of u64::MAX, so the five slots sum to exactly one. The subnet owner (or root) sets it with set-subnet-emission-weights; anyone reads it with subnet-emission-weights or btcli subnets emission show.

New subnets start at the historical split — [0, 0, 0.18, 0.41, 0.41] on mainnet — so nothing changes until an owner decides it should.

What each slot does

The owner slot at zero means "no owner cut"; the legacy owner_cut_enabled hyperparameter is now simply owner > 0. Burn and recycle are present in the vector so the format does not change when they are enabled later, but a non-zero value in either is rejected today.

Per block, alpha_out × owner accrues to the owner, alpha_out × validator to the validator bucket and alpha_out × miner to the miner bucket. Root TAO stakers are paid out of the validator bucket first; shrinking the validator slot would, on its own, shrink what they receive. The rule below prevents that.

Root off the top

Root TAO is weighted against alpha by the global TAO weight w (TaoWeight, 0.18 on mainnet today). Before emission weights existed, each block's alpha to root stakers from a subnet was

root_alpha = V × x / (x + A) × alpha_out        with x = root_tao × w

with V the fixed 41% validator slot and A the subnet's alpha issuance.

That payout is kept and detached from the owner's choice of V. Every block, for every subnet, the chain computes root's baseline share

r = V0 × x / (x + A)        with x = root_tao × TaoWeight

where V0 is the validator share of the default vector (0.41 on mainnet, derived from the default owner cut — a single global constant) — what root would earn from this subnet right now if the owner had never touched the vector. Root is paid min(V, r) × alpha_out off the top of the validator slot and alpha validators share the rest, (V − min(V, r)) × alpha_out (get_root_emission_share in emission_weights.rs, applied in run_coinbase.rs). The owner and miner slots are untouched.

Because r is recomputed from live state every block, root's payout tracks the baseline continuously — as issuance grows against root TAO, as root stake moves, or as governance changes TaoWeight — with no call from anyone. An owner lowering V from 41% to 20% takes the whole 21 points from alpha validators; root's slice is identical before and after, and no sequence of changes can ratchet it in either direction. There is no per-subnet TAO weight: stake weighting inside Yuma Consensus, the TAO-vs-alpha split of validator dividends, child/parent stake and the pool side all use the global TaoWeight exactly as before.

A set call fails with ValidatorEmissionTooLowToPreserveRoot when V_new ≤ r: root would take the entire slot and alpha validators would get nothing, so the floor at this block is r itself. On a mainnet subnet with 5.4M TAO on root, 6.5M alpha issued and TaoWeight = 0.18, r is about 5.40% of emission and the validator share can go down to anything strictly above that. The floor moves with root stake, issuance and the global weight, so btcli subnets emission show prints the current value. With no root stake there is nothing to preserve and any positive V is accepted.

The check runs at set time while the baseline keeps moving. If r later grows past a subnet's V, root's cut is capped at the whole validator slot (V × alpha_out) and alpha validators receive nothing until the owner raises V; root is never paid more than the slot. show reports this as capped, and the runtime API returns both root_baseline_share (r) and root_share (min(V, r)).

The pool side is unchanged. tao_in, alpha_in and the alpha-injection cap in get_subnet_terms use the pool-side root_proportion with the global TaoWeight, exactly as before emission weights existed. Changing a subnet's vector changes how its alpha is distributed; it does not change how much TAO or alpha is injected into its pool.

Changing the vector

Updates are sparse: name the slots you are changing and their new shares; omitted owner/validator/miner slots keep their current value. Because the five slots must still sum to one, lowering one slot means raising another in the same call. btcli does the arithmetic:

# see the vector, root's share and baseline, every allowed range, and the next change window
btcli subnets emission show --netuid 12

# set one slot; the other two are rescaled proportionally
btcli subnets emission set-value --netuid 12 --slot owner --value 0.10

# nudge one slot up or down
btcli subnets emission increase --netuid 12 --slot miner --by 5%
btcli subnets emission decrease --netuid 12 --slot validator --by 0.02

# write the whole split at once
btcli subnets emission set --netuid 12 --owner 0.10 --validator 0.40 --miner 0.50

Every write validates against the rules locally first and prints each broken rule in the same words the chain error carries, with the allowed range, so a rejected change teaches the constraint. Add --dry-run to preview the plan — the before/after vector, root's share now and after (its baseline), and what alpha validators keep — without signing; add --json for the same as a machine-readable record.

Owners may change the vector once per week per subnet (50,400 blocks). The limit is keyed by subnet, not by owner coldkey, so handing the subnet to a new owner does not reset it; the first change on a subnet is never rate-limited, root is exempt, and dissolving a subnet clears the key so a reused netuid starts fresh. Both owner and root must obey every rule, including the lease floor. Changes take effect on the next block's emission and are announced by the SubnetEmissionWeightsSet event, which carries the full post-merge vector.

Related

• Emissions — the full per-tempo flow
• owner_emission, validator_emission, miner_emission
• set-subnet-emission-weights — the intent
• subnet-emission-weights — the read