Skip to main content

Staking Rewards

Staking earns a share of the emission a Neuronet receives. The more useful work a Neuronet produces, the larger that pool, and the larger a staker's cut. This page covers what that share is, how the runtime computes each staker's part of it, and the stake thresholds that decide whether a Neuronet pays out at all.

What Stakers Earn​

Stakers receive a fixed 10% of a Neuronet's emission, divided in proportion to how much each one staked. This share is not configurable per Neuronet: it is the same 10% everywhere, set by the runtime parameter StakersSharePct.

The rest of a Neuronet's emission is split between its Owner and its miners. The Owner's cut is the Neuronet's owner fee, which defaults to 10% and can be lowered but not raised. Miners receive whatever is left after the owner fee and the staker share are taken out:

owner_share   = neuronet_emission × owner_fee / 100
staker_share  = neuronet_emission × 10 / 100
miner_share   = neuronet_emission × (100 - owner_fee - 10) / 100

With the default 10% owner fee that gives the familiar 80/10/10 split. See Emission System for how a Neuronet's emission is sized in the first place.

How a Staker's Share Is Computed​

The staker pool is not divided by raw balance. It is divided by eligible stake, which is stake that has finished maturing.

When you stake into a Neuronet, the runtime records an activation entry alongside the balance: an eligible amount and the block at which it becomes eligible. At settlement the runtime walks the stakers of that Neuronet and counts a staker only if their activation block has already passed. Stake added mid-cycle sits in the Neuronet, counts toward capacity, but does not draw a reward until it has been in place for a full epoch.

Each qualifying staker then receives:

payout = staker_share × your_eligible_stake / total_eligible_stake

Two consequences worth planning around:

  • Only eligible stake appears in the denominator, so a staker who has matured is not diluted by stake that arrived after the epoch began.
  • The arithmetic is integer multiply-then-divide, done in the chain's smallest unit. Rounding losses are at the level of individual planck and are not a practical concern, but payouts are not exact fractions.

When a Neuronet Has No Eligible Stakers​

A Neuronet can be producing verified work while having no staker who qualifies for a payout: everyone staked recently, or the Owner is the only one holding the position and holds it outside the staking path.

In that case the staker share is not burned and it is not held in reserve. The runtime treats it as orphaned and folds it into the miner pool for that settlement, so miners of that Neuronet receive the full remainder after the owner fee. Nothing is minted or destroyed by this path: the same total is paid out, just to a different set of accounts.

This matters if you are timing an entry. Staking into a Neuronet that currently has no eligible stakers does not let you collect a backlog. The share for cycles you were not eligible for has already gone to miners.

Activation: A Neuronet Needs Stake to Earn​

A Neuronet does not pay out anything until it has enough stake behind it.

  • The threshold is 100 AET of total stake on the Neuronet, set by the runtime parameter MinNeuronetActivationStake. This is the total across every staker, not a per-staker figure.
  • The Owner can stake this themselves at creation, or leave it to stakers who delegate later.
  • The check runs at settlement, not at creation. If total stake is below the threshold when a settlement boundary arrives, the runtime emits a below-threshold event and skips the payout for that cycle.

Emission earned while a Neuronet is under the threshold is held as pending rather than discarded. The runtime only clears the pending balance when it actually distributes it, so a Neuronet that crosses 100 AET later can still settle what it accrued. This is not a promise of indefinite retention: freezing, deregistration, and other status changes have their own handling, and a Neuronet parked below the threshold is also earning nothing new if it has no traffic.

Early stakers therefore play a concrete role. Their stake can be what takes a Neuronet from dormant to paying.

Stake, Capacity, and What It Actually Buys​

Stake sets the volume of work a Neuronet can be paid for. The runtime computes it like this:

staked_aet = total_staked_planck / 1_000_000_000
capacity   = staked_aet × 100 slots

PlanckPerAet is 1,000,000,000 and CapacityPerAet is 100 slots per AET. One slot is one atomic unit of work: an inference request, a shadow-replay check, or part of a training step.

The division by PlanckPerAet happens first and truncates. Fractional AET below a whole token contributes nothing to capacity, so a Neuronet with 100.9 AET staked gets the same 10,000 slots as one with exactly 100 AET.

When a batch is submitted, the runtime credits work against the remaining free capacity and silently drops the overflow, emitting a capacity-exceeded event for the part it could not credit. Work beyond capacity is still served to users, it simply earns no emission. See Staking Overview for the wider picture.

Returns Follow Real Usage​

Because emission tracks verified work, staking a busy, well-run Neuronet earns more than staking a quiet one. Adding stake to a Neuronet with no traffic mostly locks up tokens without raising earnings: capacity only pays when there is work to fill it.

There is a second reason raw size is the wrong thing to optimize. Your payout is your eligible stake divided by the Neuronet's total eligible stake, so piling stake into an already heavily staked Neuronet dilutes you against everyone else there. What raises a staker's return is the Neuronet's emission growing, and that comes from verified work being produced and passing checks. Capacity that no one fills does not generate any.

Where to Go Next​

Last updated on