Introduction

Today, every registered subnet earns emission in direct proportion to its moving price. That rule is simple but too flat: for instance, a subnet nobody uses still collects a slice of every block, which is to say, the tail of the network dilutes the subnets doing real work. Simultaneously, the passive income of a parked slot props up the cost of registering a new one, since the expected return of nothing is large. The result: the network cannot grow without taxing its champions.

Spec 440changes Bittensor's emission curve. Emission remains price-based, but prices now pass through a threshold 'gate' — an inflection point centered around the 32nd rank. Above the gate, emission is essentially unchanged. Below it, emission collapses toward zero. Technically no subnet is hard-cut — every price still gets some emission — but a parked slot pays substantially less than an active one.

Because a slot below the gate earns little, slots no longer need to be scarce — this release clears the path to raise the subnet cap later, without taxing the head today.

Thesis: a slot should cost nothing

The cost of registering a subnet has been one of the network's quietest failures. Under flat price-proportional emission, even the worst slot on the network carried a passive yield, so slots traded at inflated values. This taxed teams who wanted to build, since they needed to borrow against the dead weight of the slot itself (currently ~1300 TAO). Most took on debt to make this happen. The people the network most wants were the ones the pricing punished.

Release 440 changes this by lowering the base return of an idle slot to approximately nothing, so an idle slot is worth approximately nothing, and the price of entry should fall toward the cost of the registration transaction. In other words, a team no longer buys a guaranteed slice of emission — just a starting position below a strict competition bar. Over time, new and creative teams can earn emission by attracting substantial demand from TAO holders.

Idle yield props the market price of a slot under flat emission. With the gate, idle carry collapses and entry should fall toward the registration transaction.

The bar and the gate

We create the gate as follows: sort subnets by price and walk down until the subnets above you carry a fraction qof all prices — the price at that point is the 'bar'. Each subnet's emission weight is then thresholded around this point.

The q-mass bar: accumulate sorted demand until the running total crosses q; the share at the crossing is θ. Defaults put that crossing near rank 32 on today's Finney distribution.

// pallets/subtensor/src/coinbase/subnet_emissions.rs
let theta = Self::q_mass_bar(&shares, EmissionBarQuantile::<T>::get()); // q
let h = EmissionGateExponent::<T>::get();                               // h

// gate(s) = s^h / (s^h + theta^h); emission weight = s * gate(s)
let weights: Vec<U64F64> = shares
    .iter()
    .map(|s| {
        let sh = s.saturating_pow(h);
        s.saturating_mul(sh.safe_div(sh.saturating_add(theta.saturating_pow(h))))
    })
    .collect();

We use a smooth function for the threshold (a sigmoid). At the threshold it cuts exactly half of a subnet's emission; two ranks above, nothing is cut; deep below that rank, a subnet gets almost nothing.

Hill gate at h = 3. At s = θ the gate passes exactly ½; well above the bar it is ~1; deep in the tail it is ~0.

Notably, if the tail grows real demand, θ rises and the standard gets harder for everyone; if demand concentrates, θ falls. At the defaults — q = 0.61, h = 3— the bar lands on today's distribution at the average demand share of an active subnet, around rank 32. A subnet earns its full linear emission only if it attracts more than an average slice of demand. Because the bar is a property of the demand distribution, registering new subnets does not move it.

See the cliff

The figure below is computed from a live mainnet snapshot of all 126 subnets, with every subnet treated as emission-enabled. The muted line is emission under the old price-proportional rule; the red line is emission through the gate. Drag q to move the bar and h to sharpen or soften the cliff.

Finney SubnetMovingPrice snapshot, July 2026. Defaults q = 0.61, h = 3 — the shipped hyperparameter values.

What the defaults do

At h = 3 the cliff is decisive. A subnet just above the bar keeps its emission almost untouched; the mid-tail keeps 15–30% of its linear share; the deep tail keeps 1–5% — a token drip that marks the climb path without being worth holding. And the leverage near the bar is strong in the right direction: a subnet at rank 36 that grows its demand 10% gains roughly 26% more emission. Growth near the cliff is the best-paid move on the network.

Parameters

Both are root-sudo settable and rate-limited. The bar itself is recomputed once per tempo from the same de-manipulated moving prices that already drive emission, so a single block of wash trading cannot move it. Because a sharper h = 3 gate makes the boundary more sensitive, that per-tempo cadence is what keeps emission near the bar steady rather than flapping.

What to do

• Node operators: wait for the on-chain spec_version to move to 440, then update to the matching release.
• Subnet owners below the bar: your emission does not go to zero, but it no longer pays to idle. The bar is public — query it with btcli subnets bar and track your share against θ.
• Subnet owners above the bar:nothing to change. Your emission rises as the tail's subsidy is redistributed.
• Stakers: alpha positions in deep-tail subnets earn materially less emission after this upgrade; the head earns more. Reprice accordingly.
• Teams waiting to build: existing slots lose the passive carry that made them expensive. Expect registration cost to fall toward the transaction fee — a slot buys a starting position, not an income stream. Raising the subnet slot cap is deferred to a later release.

Signers: after the release train proposes, use btcli upgrade sign --url <v440 release URL> -w <wallet>.