Fractal subnet emission rates with emission-adjusted deregistration
Summary
Bittensor applies a common emissions framework to subnets with very different stages of development and capital requirements. A compute-heavy production subnet may require substantial emissions to operate, while an early research subnet may need time more than liquidity or scale. Requiring both to accept emissions at the same rate can create unnecessary dilution, push more TAO and alpha into subnet economies than they can productively use, and force promising research to compete against mature products on an unsuitable timeline.
This proposal introduces owner-selected fractal emission rates. A subnet owner may choose to accept 100%, 50% or 25% of its otherwise available protocol emissions. Lower powers of two, such as 12.5% and 6.25%, may be enabled later as TAO becomes more valuable and the market develops.
The result is less newly emitted TAO entering circulation, slower alpha dilution and greater runway for subnets that do not yet need full emissions.
Choosing a lower rate would also be reflected in subnet deregistration. The initial proposed model divides a subnet's moving alpha price by its selected emission rate when calculating its deregistration score. A subnet accepting 25% of its available emissions would therefore receive a 4x adjustment for deregistration purposes. This does not guarantee survival. It allows the market to evaluate the subnet over a longer period while it consumes fewer cumulative emissions.
The proposal creates a market-driven trade-off between present resources and long-term runway. Proven and capital-intensive subnets can select full emissions. Early research can select less, dilute more slowly and earn more time to demonstrate value. Subnets that accept full emissions but fail to produce useful outputs, external revenue, revenue-funded buybacks or credible progress should eventually lose investor support and become more vulnerable to deregistration.
Motivation
One emission speed does not fit every subnet
Bittensor subnets are not economically identical.
Some require large and continuous operating budgets for compute, inference, storage or other resource-intensive work. Others begin as research programmes whose most valuable output may initially be an algorithm, dataset, benchmark improvement or scientific result. These subnets may require a longer period of experimentation before they can produce revenue or support large-scale operations.
The current system largely asks both categories to develop under the same supply schedule. That creates several problems:
Early research can receive more emissions than it can productively use.
Research subnets can be forced into premature productisation simply to compete with established subnets.
Alpha issuance can outpace the market demand needed to absorb it, weakening price before the subnet has had enough time to demonstrate value.
A large rise in TAO's price can make an unchanged nominal emission rate economically excessive.
This matters because the economic value of TAO-denominated emissions can rise dramatically without any corresponding increase in what a subnet needs to operate or scale. A sharp increase in TAO's price could therefore direct far more subsidy into an already emission-heavy ecosystem than subnets can productively use. Owners should be able to scale the selected rate to match the subnet's current compute, liquidity, operating and growth requirements.
A selectable rate therefore allows each subnet to match its emissions to its actual needs instead of forcing every subnet to consume its full allocation at the same speed.
A fixed global schedule meets variable local demand
Bitcoin's issuance schedule coordinates one monetary asset and one security market. Bittensor coordinates many evolving subnet economies whose capital requirements, maturity and paths to revenue differ substantially.
Borrowing a predictable halving rhythm can still be valuable, but it does not follow that every subnet should consume its available emissions at the same speed. Fractal emission rates preserve a predictable global framework while allowing local demand for subsidy to adjust from inside the network, without relying on a TAO/USD oracle.
Reducing TAO and alpha supply pressure
Fractal rates would reduce unnecessary emission-driven supply and sell pressure at both the network and subnet levels. When a subnet cannot productively use its full allocation, selecting a lower rate means less newly emitted TAO enters circulation and fewer alpha tokens are issued.
Declined TAO would be burned rather than redistributed, reducing circulating-supply growth without altering the existing TAO halving path. Within the subnet, slower alpha issuance reduces dilution and the supply available to be sold. Subnets that can productively use full funding remain free to select 100%.
Less emission, more runway
Holding all other factors constant, a subnet selecting 25% would consume approximately the same cumulative emission budget over four periods that it would consume in one period at 100%.
It is therefore reasonable for the lower-rate subnet's emission choice to be considered in deregistration. Otherwise, the subnet receives only one-quarter of the resources but must compete for survival using the same unadjusted moving alpha price as a subnet accepting four times the subsidy.
The adjusted score does not guarantee a fixed extension of calendar time. It makes deregistration account for the proportion of available emissions each subnet chooses to accept.
Selectable rates could enable a softer emission gate
Allowing subnets to limit how much they accept could make a softer emission gate possible, giving emerging projects a more usable allocation without forcing them to consume it all. This would give early research enough support to operate while keeping its subsidy proportionate to its stage of development. Bittensor could support a wider field of experiments without treating each one as though it already needs production-scale emissions.
A recent example: Babelbit
Babelbit's SN59 was deregistered roughly one week after the team announced the Babelbit Dubbing API, the first product built on its Language Transformation infrastructure. The announcement followed a major multilingual-model breakthrough shared several days earlier and described a programmable real-time dubbing product for broadcasters, media companies, voice agents and developers.
The timing makes Babelbit an unfortunate example of why research subnets may benefit from an adaptable emission rate. Under the proposed linear model, selecting 50% during an earlier research phase would have reduced the subnet's owner, miner and validator/staker rewards in exchange for slower alpha dilution and a 2x deregistration adjustment. Depending on the surrounding rankings, that could have given Babelbit months or longer to convert its progress into adoption and market support, with the owner able to increase or decrease the rate as its needs changed.
Babelbit had fallen into the near-zero-emission tail before deregistration. Under a softer gate curve, 50% of a larger pre-rate allocation could still have exceeded what it received at 100% under the current gate. With lower liquidity, similar investor support could also have had a greater effect on Babelbit's alpha price, potentially improving the emission allocation it earned before the 50% rate was applied. These effects would depend on market conditions and the exact gate design, but Babelbit's experience provides a concrete test case for whether productive subnets can trade short-term rewards for more time and a better chance to mature.
Proposed mechanism
1. Owner-selected emission rates
Each subnet owner selects one of three initial emission rates:
The rate set may later be extended fractally to 12.5%, 6.25% and lower levels if the network determines that they are useful. Starting with 100%, 50% and 25% limits complexity while producing enough variation to test the mechanism.
Existing subnets would default to 100%. No subnet would be placed into a reduced mode without an explicit owner decision.
2. Accepted emissions
Market support and the emission gate first determine the subnet's TAO allocation. The selected rate then scales how much of that TAO allocation is accepted and how quickly new alpha tokens are issued for the subnet. A subnet selecting 50% accepts half of its otherwise available protocol emissions. A subnet selecting 25% accepts one-quarter.
Emissions the subnet does not accept should not be redistributed to other subnets as additional emissions. Redistribution would keep the same amount of newly emitted TAO entering circulation and remove the central supply benefit of the proposal.
The preferred treatment is to burn declined TAO. Under current Bittensor accounting, burned TAO remains counted towards total issuance and halving thresholds. This preserves the existing TAO halving path while preventing declined emissions from entering circulation or being redistributed to other subnets.
Leaving TAO unissued or recycling it remain possible implementation alternatives, but neither is proposed as the base treatment. Leaving it unissued could extend the overall TAO emission timeline, while recycling it would allow the TAO to be emitted again later.
3. Emission-adjusted deregistration
The starting proposal divides the subnet's moving alpha price by its selected emission rate to produce an emission-adjusted deregistration score. Under this linear model:
Selected rate |
Deregistration multiplier |
100% |
1x |
50% |
2x |
25% |
4x |
12.5% |
8x |
6.25% |
16x |
For example, consider the same subnet while holding its raw moving alpha price and pre-rate allocation constant:
At 100%, a moving alpha price of 0.01 produces an adjusted deregistration score of 0.01, and the subnet accepts its full per-period emissions.
At 25%, the same moving alpha price produces an adjusted deregistration score of 0.04, and the subnet accepts one-quarter of those emissions per period.
If the subnet's underlying allocation and other emission conditions remain unchanged, four periods at 25% provide the same cumulative subsidy as one period at 100%. This is the runway benefit: the market receives more calendar time to observe progress without granting the subnet a larger cumulative subsidy.
This adjustment does not provide fixed or guaranteed immunity. If a 25% subnet loses nearly all market support, four times a near-zero price remains near zero. Once its immunity has ended, it can still be deregistered if it has the network's lowest adjusted score when a new subnet registers.
The existing network-immunity period should initially remain unchanged. The selected rate would affect the deregistration score after immunity, not extend hard immunity by a fixed number of months.
4. Rate changes and maturation
A subnet may change its selected rate as its needs change.
A typical research-to-product path might be:
Launch at 25% while conducting early research.
Publish results, benchmarks, code or other verifiable progress.
Attract greater market support as the thesis is validated.
Move to 50% when more miners, validators, liquidity or operating resources become useful.
Move to 100% when the subnet has a proven product, significant operating costs or a credible path towards revenue and buybacks.
The inverse path should also be possible. A mature subnet whose capital requirements decline could voluntarily reduce its rate, slow dilution and reduce the TAO entering circulation.
To prevent rate switching from becoming a short-term tactic rather than a genuine economic commitment:
An increase in emissions should remove the associated deregistration advantage immediately.
A reduction in emissions should reduce accepted emissions immediately, while its additional deregistration adjustment should become active gradually.
A cooldown should prevent owners from repeatedly switching between a lower rate for deregistration protection and a higher rate for rewards. It would also give miners, validators, stakers and investors time to adjust to the subnet's new budget.
Because rate reductions already cut emissions immediately and earn their deregistration benefit gradually, the cooldown may not need to be long. Its duration should be determined through simulation and technical review.
5. Consistent economic scaling
Selecting 25% must create a genuine reduction in the subnet's protocol subsidy rather than merely moving emissions from one destination to another.
The intended properties are:
The selected rate applies to both the TAO and alpha emissions otherwise made available to or through the subnet, rather than reducing only one reward category.
A subnet selecting 25% receives 25% of its otherwise available owner, miner and validator/staker reward pools, along with 25% of its otherwise available protocol-funded chain buying.
Lower selected rates slow alpha issuance, alpha dilution and the subnet's alpha-halving timeline, providing more calendar runway per unit of protocol subsidy consumed.
Higher selected rates provide larger reward pools and more protocol funding for liquidity injection, chain buying, infrastructure and operations appropriate for scaling a proven or capital-intensive subnet.
Emissions the subnet declines are not redistributed to other subnets or recreated through a different form of current protocol subsidy.
Protocol engineers should determine how existing mechanisms are modified so that the selected rate is applied consistently across the system.
Protocol-funded chain buying should scale with the selected rate because it remains part of the subnet's protocol subsidy. External revenue and revenue-funded buybacks should remain unaffected because they represent genuine economic demand rather than emissions. At a lower selected rate, the same external revenue or buyback also becomes more meaningful relative to the subnet's reduced protocol subsidy and alpha issuance.
Game theory and market discipline
This proposal is a mechanism-design change. It gives subnet owners a choice, but attaches visible economic consequences to each choice.
The full-emission strategy
A subnet selecting 100% receives the maximum resources available under its market-earned allocation. That can support more compute, stronger miner and validator incentives, deeper liquidity, broader operations and faster scaling.
However, it also accepts the fastest dilution and receives no deregistration adjustment. The market should therefore expect correspondingly greater delivery.
If a full-rate subnet continually consumes emissions but fails to produce useful outputs, users, external revenue, revenue-funded buybacks or credible progress, several effects should follow:
New emissions create continuing supply and potential sell pressure.
The absence of external revenue or revenue-funded buybacks leaves less organic demand to absorb that supply.
Investors revise their expectations and reduce exposure.
The subnet's moving alpha price weakens.
Lower-rate subnets producing more progress per unit of emission can achieve a safer adjusted deregistration score.
At any stage, an owner is incentivised to reduce the subnet's emission rate when its full allocation exceeds what it can productively use. As market support weakens, this becomes an escape route from deregistration: the subnet gives up excess rewards today for greater runway and an opportunity to rebuild support.
If the subnet refuses to adapt and continues consuming full emissions without delivering, it eventually becomes a candidate for deregistration.
Full emissions therefore stop being the automatic safest option. They become a claim that the subnet can productively use the maximum available subsidy.
The reduced-emission strategy
For early and unproven research that does not yet need production-scale funding or liquidity, selecting 50% or 25% can be appropriate. Investors are not asked to fund full dilution before sufficient evidence exists.
In exchange, each unit of moving-price support contributes more to the subnet's adjusted deregistration score. When a new subnet registers at capacity, the non-immune subnet with the lowest adjusted score would be deregistered.
A research subnet that never produces meaningful progress should eventually lose even the limited investor support required to maintain its adjusted score. By the time the market reaches that conclusion, the network would have directed substantially fewer emissions to it than if it had remained at 100%.
The trade-off cuts both ways:
A subnet selecting 25% gives up full-rate miner budgets, liquidity and operating capacity, but gains longer runway and slower alpha dilution.
A resource-intensive subnet selecting 100% receives maximum funding, but without sufficient demand to absorb its emissions, the resulting higher sell pressure may eventually weaken its alpha price and market support.
Graduation rather than permanent classification
The protocol should not attempt to decide which subnets are "research subnets" and which are "production subnets". Such labels would be subjective and easy to game.
The emission rate itself expresses the subnet's stage and capital requirements. A subnet may begin at 25%, demonstrate value and graduate towards 100%. Another may remain at 25% because research is its enduring product and its operating requirements remain low.
The market judges whether the selected rate is credible through price, continued backing and eventual evidence of value creation.
Rationale
Why powers of two?
Powers of two create a simple and legible relationship between emissions and deregistration treatment:
Half the emissions, twice the deregistration-score multiplier.
One-quarter of the emissions, four times the multiplier.
One-eighth of the emissions, eight times the multiplier.
The pattern is easy for owners, investors and protocol participants to understand. It can also extend as TAO's value and the subnet economy grow without introducing a TAO/USD oracle or frequent governance intervention.
Why begin with a linear adjustment?
The linear formula directly expresses the proposal's central principle: deregistration should consider market support relative to the proportion of emissions accepted.
A softer curve could reduce the ranking benefit at very low rates if testing shows that the linear model makes subnet slots too inexpensive to maintain. However, the linear version is the clearest hypothesis and should be tested before additional complexity is introduced.
Why owner selection?
Subnet owners understand their operating requirements and development stage better than network-wide governance. They are also directly exposed to the consequences of choosing an unsuitable rate.
The choice remains visible to investors. An owner cannot quietly receive the benefits of a research-rate subnet while accepting full emissions.
Why not create a separate research category?
A formal research classification would require governance or an oracle to determine what qualifies as research, whether sufficient progress is being made and when a subnet becomes a product. This would introduce subjective judgement and invite category gaming.
Selectable rates allow the subnet to reveal its own needs while the market evaluates the outcome.
Why not redistribute the unaccepted emissions?
Redistribution would improve the relative position of full-rate subnets but would leave the total amount of new TAO entering circulation unchanged. Burning declined TAO instead reduces the amount entering circulation when subnets cannot productively use their full allocation.
The strongest subnets already receive preferential treatment through market allocation and the emission gate. They should not automatically receive every emission declined by earlier-stage subnets.
Why not compensate reduced-rate investors with additional alpha?
A lower selected rate should not be offset with bonus alpha. Its benefits already come from slower dilution, a longer potential reward runway and the emission-adjusted deregistration score. Minting bonus alpha would recreate part of the dilution and rewards that the lower rate is meant to reduce.
Interaction with the existing emission gate
The emission gate creates a steep allocation curve based on market support. Highly supported subnets receive meaningful allocations; the lower-ranked tail receives very little. Fractal emission rates add a second decision:
The emission gate determines the subnet's market-earned allocation.
The selected rate then scales all of the subnet's protocol-funded emissions, including participant rewards, liquidity support and chain buying.
Under current rules, every eligible subnet automatically receives its final gate-adjusted allocation; there is no owner choice to accept less. The gate therefore needs a steep curve below its threshold to prevent the lower-ranked tail from receiving too much of the emission budget.
If emerging subnets can voluntarily select 50% or 25%, the gate may be able to use a softer curve while keeping total accepted emissions restrained. This matters because a subnet whose allocation has already been reduced to almost nothing by the gate is unlikely to volunteer for another reduction. A softer gate can leave emerging subnets with enough of a pre-rate allocation that selecting 50% or 25% remains a useful strategic choice instead of making 100% the only rational option.
The gate and fractal rates should therefore not be designed independently. Their combined effect should preserve meaningful competition throughout the ranking while preventing the tail from draining excessive TAO.
Additional subnet slots and a broader discovery market
Expanding the total number of subnet slots would complement fractal rates by reducing the opportunity cost of letting more experiments develop in parallel. More slots would widen the discovery market, reduce artificial scarcity and give more genuine projects room to compete for market support.
This combination could turn subnet slots into a broader discovery market:
The head contains proven, strongly supported subnets. They can justify higher selected rates and receive the largest allocations to fund scale, infrastructure and operations.
The middle contains emerging subnets with enough market-earned allocation to make 50% or 25% a viable choice. They compete to demonstrate progress and attract support without immediately consuming full emissions.
The tail contains early or weakly supported experiments. The gate limits the allocation they earn, and a lower selected rate can reduce what they consume even further. If progress attracts demand, they can move upwards; if it does not, they remain in the tail, exposed to deregistration when a new subnet registers if they hold the lowest non-immune adjusted score.
A long-lived experiment is not necessarily wasteful if it consumes negligible emissions and remains exposed to market-based deregistration whenever a new subnet registers at capacity.
Backwards compatibility
All existing subnets would begin at the 100% default rate, so the selectable-rate mechanism alone would not change their initial economic behaviour or deregistration score. However, the required adjustment to the emission-gate curve could change subnet allocations even at 100%. The combined effects should therefore be modelled and tested before the two mechanisms are introduced together.
The proposal requires changes to core emission and subnet-deregistration logic. The protocol should expose enough information for explorers to display the subnet's selected rate, raw moving alpha price, adjusted deregistration score and rate-change cooldown. Where technically possible, explorers should also show the estimated full-rate allocation, accepted emissions and emissions not accepted.
Security considerations
Last-minute rate switching
Without delays, a weak subnet could select 25% immediately before deregistration to obtain a 4x deregistration multiplier. Rate reductions should therefore reduce emissions immediately but grant the additional deregistration adjustment gradually. Rate increases should remove the prior advantage immediately.
Low-liquidity price manipulation
Lower liquidity can make prices easier to move in either direction. However, a subnet selecting a lower rate also receives fewer emissions, so manipulating its adjusted deregistration position does not automatically create a profitable emission-extraction strategy. Under a linear adjustment, an attacker seeking maximum emissions may be better served selecting 100% and overstating revenue, growth or progress than voluntarily accepting a lower rate.
The more relevant question is whether a low-rate subnet could maintain a slot too cheaply. The moving-price calculation, reduced emissions, ongoing market scrutiny and expansion of the total number of subnet slots all limit this risk, although it remains worth testing.
Slot warehousing
A low-rate subnet attempting to warehouse a slot extracts emissions much more slowly, giving the market more time to determine that it is not building anything. The emission gate further reduces what an unsupported tail subnet can extract. Expanding the total number of subnet slots makes each slot cheaper, allows more real projects to compete and makes hoarding any individual slot less valuable.
Historical simulation should still estimate the cost of maintaining a weak subnet at 25%, 12.5% or 6.25% under both linear and softer deregistration adjustments.
Owner and investor incentive differences
Owners choose the rate, while investors and subnet participants bear part of its consequences. Transparent interfaces, change delays and predictable cooldowns are therefore essential. A subnet whose owner repeatedly selects an unsuitable rate should lose market support.
Insufficient operating rewards
A resource-intensive subnet may select too little emission to operate effectively. If it cannot produce its commodity at the selected rate, it must increase the rate, attract outside revenue or improve its efficiency.
Open technical questions
This proposal defines the intended economic behaviour more precisely than the code changes needed to implement it. Technical contributors are requested to help determine:
How should the emission-gate curve be adjusted so that 50% and 25% remain viable choices for lower-ranked subnets while explicitly controlling the resulting effect on leading-subnet allocations and the tail's total accepted emissions?
What protocol treatment is required to burn declined TAO while preserving current total-issuance and halving accounting?
Should halving the selected rate always double the deregistration adjustment, or should the benefit increase more slowly at very low rates?
How long should the rate-change cooldown and gradual activation of the additional deregistration adjustment last?
What should the initial minimum rate be, and what evidence should justify enabling 12.5%, 6.25% or lower modes?
Could a low-rate subnet maintain an adjusted moving price and occupy a slot too cheaply, even though it receives fewer emissions?
What protocol changes are needed to apply the selected percentage consistently across all existing protocol-funded emissions, so that no emission route remains at 100% by mistake?
If this system had existed previously, how much TAO would have been kept out of circulation, how much alpha issuance would have been avoided, which subnets would have survived or been deregistered, and how cheaply could weak subnets have maintained slots?
These questions should be resolved through community discussion, historical modelling and testnet implementation before a final specification is submitted.
Evaluation criteria
A successful implementation should demonstrate that:
Reduced-rate subnets genuinely reduce the amount of newly emitted TAO entering circulation and reduce actual alpha issuance.
Research and early-stage subnets receive more evaluation time per unit of emission consumed.
Full-rate subnets remain able to fund expensive operations and scale proven products.
Full-rate subnets that fail to create value face greater market and deregistration pressure.
Proven lower-rate subnets can graduate smoothly towards higher rates.
Unaccepted emissions are not silently redistributed in a way that recreates the original supply pressure.
Low-rate modes do not make price manipulation or slot warehousing economically trivial.
The mechanism remains understandable to owners, investors, miners and validators.
Testing and implementation
No implementation is included at this stage.
The recommended next step is a simulation using historical moving alpha prices, subnet emission allocations and deregistration events. The simulation should compare:
Existing rules.
The linear model, where halving the selected emission rate doubles the deregistration adjustment.
Softer models, where the adjustment grows more slowly at very low rates.
Different cooldown periods and schedules for gradually activating the deregistration adjustment.
The current emission gate, softer gate curves and no emission gate, each tested with selected rates.
Different total numbers of subnet slots.
Initial rates of 100%, 50% and 25%, followed by possible 12.5% and 6.25% modes.
References
Fractal subnet emission rates with emission-adjusted deregistration
Summary
Bittensor applies a common emissions framework to subnets with very different stages of development and capital requirements. A compute-heavy production subnet may require substantial emissions to operate, while an early research subnet may need time more than liquidity or scale. Requiring both to accept emissions at the same rate can create unnecessary dilution, push more TAO and alpha into subnet economies than they can productively use, and force promising research to compete against mature products on an unsuitable timeline.
This proposal introduces owner-selected fractal emission rates. A subnet owner may choose to accept 100%, 50% or 25% of its otherwise available protocol emissions. Lower powers of two, such as 12.5% and 6.25%, may be enabled later as TAO becomes more valuable and the market develops.
The result is less newly emitted TAO entering circulation, slower alpha dilution and greater runway for subnets that do not yet need full emissions.
Choosing a lower rate would also be reflected in subnet deregistration. The initial proposed model divides a subnet's moving alpha price by its selected emission rate when calculating its deregistration score. A subnet accepting 25% of its available emissions would therefore receive a 4x adjustment for deregistration purposes. This does not guarantee survival. It allows the market to evaluate the subnet over a longer period while it consumes fewer cumulative emissions.
The proposal creates a market-driven trade-off between present resources and long-term runway. Proven and capital-intensive subnets can select full emissions. Early research can select less, dilute more slowly and earn more time to demonstrate value. Subnets that accept full emissions but fail to produce useful outputs, external revenue, revenue-funded buybacks or credible progress should eventually lose investor support and become more vulnerable to deregistration.
Motivation
One emission speed does not fit every subnet
Bittensor subnets are not economically identical.
Some require large and continuous operating budgets for compute, inference, storage or other resource-intensive work. Others begin as research programmes whose most valuable output may initially be an algorithm, dataset, benchmark improvement or scientific result. These subnets may require a longer period of experimentation before they can produce revenue or support large-scale operations.
The current system largely asks both categories to develop under the same supply schedule. That creates several problems:
Early research can receive more emissions than it can productively use.
Research subnets can be forced into premature productisation simply to compete with established subnets.
Alpha issuance can outpace the market demand needed to absorb it, weakening price before the subnet has had enough time to demonstrate value.
A large rise in TAO's price can make an unchanged nominal emission rate economically excessive.
This matters because the economic value of TAO-denominated emissions can rise dramatically without any corresponding increase in what a subnet needs to operate or scale. A sharp increase in TAO's price could therefore direct far more subsidy into an already emission-heavy ecosystem than subnets can productively use. Owners should be able to scale the selected rate to match the subnet's current compute, liquidity, operating and growth requirements.
A selectable rate therefore allows each subnet to match its emissions to its actual needs instead of forcing every subnet to consume its full allocation at the same speed.
A fixed global schedule meets variable local demand
Bitcoin's issuance schedule coordinates one monetary asset and one security market. Bittensor coordinates many evolving subnet economies whose capital requirements, maturity and paths to revenue differ substantially.
Borrowing a predictable halving rhythm can still be valuable, but it does not follow that every subnet should consume its available emissions at the same speed. Fractal emission rates preserve a predictable global framework while allowing local demand for subsidy to adjust from inside the network, without relying on a TAO/USD oracle.
Reducing TAO and alpha supply pressure
Fractal rates would reduce unnecessary emission-driven supply and sell pressure at both the network and subnet levels. When a subnet cannot productively use its full allocation, selecting a lower rate means less newly emitted TAO enters circulation and fewer alpha tokens are issued.
Declined TAO would be burned rather than redistributed, reducing circulating-supply growth without altering the existing TAO halving path. Within the subnet, slower alpha issuance reduces dilution and the supply available to be sold. Subnets that can productively use full funding remain free to select 100%.
Less emission, more runway
Holding all other factors constant, a subnet selecting 25% would consume approximately the same cumulative emission budget over four periods that it would consume in one period at 100%.
It is therefore reasonable for the lower-rate subnet's emission choice to be considered in deregistration. Otherwise, the subnet receives only one-quarter of the resources but must compete for survival using the same unadjusted moving alpha price as a subnet accepting four times the subsidy.
The adjusted score does not guarantee a fixed extension of calendar time. It makes deregistration account for the proportion of available emissions each subnet chooses to accept.
Selectable rates could enable a softer emission gate
Allowing subnets to limit how much they accept could make a softer emission gate possible, giving emerging projects a more usable allocation without forcing them to consume it all. This would give early research enough support to operate while keeping its subsidy proportionate to its stage of development. Bittensor could support a wider field of experiments without treating each one as though it already needs production-scale emissions.
A recent example: Babelbit
Babelbit's SN59 was deregistered roughly one week after the team announced the Babelbit Dubbing API, the first product built on its Language Transformation infrastructure. The announcement followed a major multilingual-model breakthrough shared several days earlier and described a programmable real-time dubbing product for broadcasters, media companies, voice agents and developers.
The timing makes Babelbit an unfortunate example of why research subnets may benefit from an adaptable emission rate. Under the proposed linear model, selecting 50% during an earlier research phase would have reduced the subnet's owner, miner and validator/staker rewards in exchange for slower alpha dilution and a 2x deregistration adjustment. Depending on the surrounding rankings, that could have given Babelbit months or longer to convert its progress into adoption and market support, with the owner able to increase or decrease the rate as its needs changed.
Babelbit had fallen into the near-zero-emission tail before deregistration. Under a softer gate curve, 50% of a larger pre-rate allocation could still have exceeded what it received at 100% under the current gate. With lower liquidity, similar investor support could also have had a greater effect on Babelbit's alpha price, potentially improving the emission allocation it earned before the 50% rate was applied. These effects would depend on market conditions and the exact gate design, but Babelbit's experience provides a concrete test case for whether productive subnets can trade short-term rewards for more time and a better chance to mature.
Proposed mechanism
1. Owner-selected emission rates
Each subnet owner selects one of three initial emission rates:
100%
50%
25%
The rate set may later be extended fractally to 12.5%, 6.25% and lower levels if the network determines that they are useful. Starting with 100%, 50% and 25% limits complexity while producing enough variation to test the mechanism.
Existing subnets would default to 100%. No subnet would be placed into a reduced mode without an explicit owner decision.
2. Accepted emissions
Market support and the emission gate first determine the subnet's TAO allocation. The selected rate then scales how much of that TAO allocation is accepted and how quickly new alpha tokens are issued for the subnet. A subnet selecting 50% accepts half of its otherwise available protocol emissions. A subnet selecting 25% accepts one-quarter.
Emissions the subnet does not accept should not be redistributed to other subnets as additional emissions. Redistribution would keep the same amount of newly emitted TAO entering circulation and remove the central supply benefit of the proposal.
The preferred treatment is to burn declined TAO. Under current Bittensor accounting, burned TAO remains counted towards total issuance and halving thresholds. This preserves the existing TAO halving path while preventing declined emissions from entering circulation or being redistributed to other subnets.
Leaving TAO unissued or recycling it remain possible implementation alternatives, but neither is proposed as the base treatment. Leaving it unissued could extend the overall TAO emission timeline, while recycling it would allow the TAO to be emitted again later.
3. Emission-adjusted deregistration
The starting proposal divides the subnet's moving alpha price by its selected emission rate to produce an emission-adjusted deregistration score. Under this linear model:
Selected rate
Deregistration multiplier
100%
1x
50%
2x
25%
4x
12.5%
8x
6.25%
16x
For example, consider the same subnet while holding its raw moving alpha price and pre-rate allocation constant:
At 100%, a moving alpha price of 0.01 produces an adjusted deregistration score of 0.01, and the subnet accepts its full per-period emissions.
At 25%, the same moving alpha price produces an adjusted deregistration score of 0.04, and the subnet accepts one-quarter of those emissions per period.
If the subnet's underlying allocation and other emission conditions remain unchanged, four periods at 25% provide the same cumulative subsidy as one period at 100%. This is the runway benefit: the market receives more calendar time to observe progress without granting the subnet a larger cumulative subsidy.
This adjustment does not provide fixed or guaranteed immunity. If a 25% subnet loses nearly all market support, four times a near-zero price remains near zero. Once its immunity has ended, it can still be deregistered if it has the network's lowest adjusted score when a new subnet registers.
The existing network-immunity period should initially remain unchanged. The selected rate would affect the deregistration score after immunity, not extend hard immunity by a fixed number of months.
4. Rate changes and maturation
A subnet may change its selected rate as its needs change.
A typical research-to-product path might be:
Launch at 25% while conducting early research.
Publish results, benchmarks, code or other verifiable progress.
Attract greater market support as the thesis is validated.
Move to 50% when more miners, validators, liquidity or operating resources become useful.
Move to 100% when the subnet has a proven product, significant operating costs or a credible path towards revenue and buybacks.
The inverse path should also be possible. A mature subnet whose capital requirements decline could voluntarily reduce its rate, slow dilution and reduce the TAO entering circulation.
To prevent rate switching from becoming a short-term tactic rather than a genuine economic commitment:
An increase in emissions should remove the associated deregistration advantage immediately.
A reduction in emissions should reduce accepted emissions immediately, while its additional deregistration adjustment should become active gradually.
A cooldown should prevent owners from repeatedly switching between a lower rate for deregistration protection and a higher rate for rewards. It would also give miners, validators, stakers and investors time to adjust to the subnet's new budget.
Because rate reductions already cut emissions immediately and earn their deregistration benefit gradually, the cooldown may not need to be long. Its duration should be determined through simulation and technical review.
5. Consistent economic scaling
Selecting 25% must create a genuine reduction in the subnet's protocol subsidy rather than merely moving emissions from one destination to another.
The intended properties are:
The selected rate applies to both the TAO and alpha emissions otherwise made available to or through the subnet, rather than reducing only one reward category.
A subnet selecting 25% receives 25% of its otherwise available owner, miner and validator/staker reward pools, along with 25% of its otherwise available protocol-funded chain buying.
Lower selected rates slow alpha issuance, alpha dilution and the subnet's alpha-halving timeline, providing more calendar runway per unit of protocol subsidy consumed.
Higher selected rates provide larger reward pools and more protocol funding for liquidity injection, chain buying, infrastructure and operations appropriate for scaling a proven or capital-intensive subnet.
Emissions the subnet declines are not redistributed to other subnets or recreated through a different form of current protocol subsidy.
Protocol engineers should determine how existing mechanisms are modified so that the selected rate is applied consistently across the system.
Protocol-funded chain buying should scale with the selected rate because it remains part of the subnet's protocol subsidy. External revenue and revenue-funded buybacks should remain unaffected because they represent genuine economic demand rather than emissions. At a lower selected rate, the same external revenue or buyback also becomes more meaningful relative to the subnet's reduced protocol subsidy and alpha issuance.
Game theory and market discipline
This proposal is a mechanism-design change. It gives subnet owners a choice, but attaches visible economic consequences to each choice.
The full-emission strategy
A subnet selecting 100% receives the maximum resources available under its market-earned allocation. That can support more compute, stronger miner and validator incentives, deeper liquidity, broader operations and faster scaling.
However, it also accepts the fastest dilution and receives no deregistration adjustment. The market should therefore expect correspondingly greater delivery.
If a full-rate subnet continually consumes emissions but fails to produce useful outputs, users, external revenue, revenue-funded buybacks or credible progress, several effects should follow:
New emissions create continuing supply and potential sell pressure.
The absence of external revenue or revenue-funded buybacks leaves less organic demand to absorb that supply.
Investors revise their expectations and reduce exposure.
The subnet's moving alpha price weakens.
Lower-rate subnets producing more progress per unit of emission can achieve a safer adjusted deregistration score.
At any stage, an owner is incentivised to reduce the subnet's emission rate when its full allocation exceeds what it can productively use. As market support weakens, this becomes an escape route from deregistration: the subnet gives up excess rewards today for greater runway and an opportunity to rebuild support.
If the subnet refuses to adapt and continues consuming full emissions without delivering, it eventually becomes a candidate for deregistration.
Full emissions therefore stop being the automatic safest option. They become a claim that the subnet can productively use the maximum available subsidy.
The reduced-emission strategy
For early and unproven research that does not yet need production-scale funding or liquidity, selecting 50% or 25% can be appropriate. Investors are not asked to fund full dilution before sufficient evidence exists.
In exchange, each unit of moving-price support contributes more to the subnet's adjusted deregistration score. When a new subnet registers at capacity, the non-immune subnet with the lowest adjusted score would be deregistered.
A research subnet that never produces meaningful progress should eventually lose even the limited investor support required to maintain its adjusted score. By the time the market reaches that conclusion, the network would have directed substantially fewer emissions to it than if it had remained at 100%.
The trade-off cuts both ways:
A subnet selecting 25% gives up full-rate miner budgets, liquidity and operating capacity, but gains longer runway and slower alpha dilution.
A resource-intensive subnet selecting 100% receives maximum funding, but without sufficient demand to absorb its emissions, the resulting higher sell pressure may eventually weaken its alpha price and market support.
Graduation rather than permanent classification
The protocol should not attempt to decide which subnets are "research subnets" and which are "production subnets". Such labels would be subjective and easy to game.
The emission rate itself expresses the subnet's stage and capital requirements. A subnet may begin at 25%, demonstrate value and graduate towards 100%. Another may remain at 25% because research is its enduring product and its operating requirements remain low.
The market judges whether the selected rate is credible through price, continued backing and eventual evidence of value creation.
Rationale
Why powers of two?
Powers of two create a simple and legible relationship between emissions and deregistration treatment:
Half the emissions, twice the deregistration-score multiplier.
One-quarter of the emissions, four times the multiplier.
One-eighth of the emissions, eight times the multiplier.
The pattern is easy for owners, investors and protocol participants to understand. It can also extend as TAO's value and the subnet economy grow without introducing a TAO/USD oracle or frequent governance intervention.
Why begin with a linear adjustment?
The linear formula directly expresses the proposal's central principle: deregistration should consider market support relative to the proportion of emissions accepted.
A softer curve could reduce the ranking benefit at very low rates if testing shows that the linear model makes subnet slots too inexpensive to maintain. However, the linear version is the clearest hypothesis and should be tested before additional complexity is introduced.
Why owner selection?
Subnet owners understand their operating requirements and development stage better than network-wide governance. They are also directly exposed to the consequences of choosing an unsuitable rate.
The choice remains visible to investors. An owner cannot quietly receive the benefits of a research-rate subnet while accepting full emissions.
Why not create a separate research category?
A formal research classification would require governance or an oracle to determine what qualifies as research, whether sufficient progress is being made and when a subnet becomes a product. This would introduce subjective judgement and invite category gaming.
Selectable rates allow the subnet to reveal its own needs while the market evaluates the outcome.
Why not redistribute the unaccepted emissions?
Redistribution would improve the relative position of full-rate subnets but would leave the total amount of new TAO entering circulation unchanged. Burning declined TAO instead reduces the amount entering circulation when subnets cannot productively use their full allocation.
The strongest subnets already receive preferential treatment through market allocation and the emission gate. They should not automatically receive every emission declined by earlier-stage subnets.
Why not compensate reduced-rate investors with additional alpha?
A lower selected rate should not be offset with bonus alpha. Its benefits already come from slower dilution, a longer potential reward runway and the emission-adjusted deregistration score. Minting bonus alpha would recreate part of the dilution and rewards that the lower rate is meant to reduce.
Interaction with the existing emission gate
The emission gate creates a steep allocation curve based on market support. Highly supported subnets receive meaningful allocations; the lower-ranked tail receives very little. Fractal emission rates add a second decision:
The emission gate determines the subnet's market-earned allocation.
The selected rate then scales all of the subnet's protocol-funded emissions, including participant rewards, liquidity support and chain buying.
Under current rules, every eligible subnet automatically receives its final gate-adjusted allocation; there is no owner choice to accept less. The gate therefore needs a steep curve below its threshold to prevent the lower-ranked tail from receiving too much of the emission budget.
If emerging subnets can voluntarily select 50% or 25%, the gate may be able to use a softer curve while keeping total accepted emissions restrained. This matters because a subnet whose allocation has already been reduced to almost nothing by the gate is unlikely to volunteer for another reduction. A softer gate can leave emerging subnets with enough of a pre-rate allocation that selecting 50% or 25% remains a useful strategic choice instead of making 100% the only rational option.
The gate and fractal rates should therefore not be designed independently. Their combined effect should preserve meaningful competition throughout the ranking while preventing the tail from draining excessive TAO.
Additional subnet slots and a broader discovery market
Expanding the total number of subnet slots would complement fractal rates by reducing the opportunity cost of letting more experiments develop in parallel. More slots would widen the discovery market, reduce artificial scarcity and give more genuine projects room to compete for market support.
This combination could turn subnet slots into a broader discovery market:
The head contains proven, strongly supported subnets. They can justify higher selected rates and receive the largest allocations to fund scale, infrastructure and operations.
The middle contains emerging subnets with enough market-earned allocation to make 50% or 25% a viable choice. They compete to demonstrate progress and attract support without immediately consuming full emissions.
The tail contains early or weakly supported experiments. The gate limits the allocation they earn, and a lower selected rate can reduce what they consume even further. If progress attracts demand, they can move upwards; if it does not, they remain in the tail, exposed to deregistration when a new subnet registers if they hold the lowest non-immune adjusted score.
A long-lived experiment is not necessarily wasteful if it consumes negligible emissions and remains exposed to market-based deregistration whenever a new subnet registers at capacity.
Backwards compatibility
All existing subnets would begin at the 100% default rate, so the selectable-rate mechanism alone would not change their initial economic behaviour or deregistration score. However, the required adjustment to the emission-gate curve could change subnet allocations even at 100%. The combined effects should therefore be modelled and tested before the two mechanisms are introduced together.
The proposal requires changes to core emission and subnet-deregistration logic. The protocol should expose enough information for explorers to display the subnet's selected rate, raw moving alpha price, adjusted deregistration score and rate-change cooldown. Where technically possible, explorers should also show the estimated full-rate allocation, accepted emissions and emissions not accepted.
Security considerations
Last-minute rate switching
Without delays, a weak subnet could select 25% immediately before deregistration to obtain a 4x deregistration multiplier. Rate reductions should therefore reduce emissions immediately but grant the additional deregistration adjustment gradually. Rate increases should remove the prior advantage immediately.
Low-liquidity price manipulation
Lower liquidity can make prices easier to move in either direction. However, a subnet selecting a lower rate also receives fewer emissions, so manipulating its adjusted deregistration position does not automatically create a profitable emission-extraction strategy. Under a linear adjustment, an attacker seeking maximum emissions may be better served selecting 100% and overstating revenue, growth or progress than voluntarily accepting a lower rate.
The more relevant question is whether a low-rate subnet could maintain a slot too cheaply. The moving-price calculation, reduced emissions, ongoing market scrutiny and expansion of the total number of subnet slots all limit this risk, although it remains worth testing.
Slot warehousing
A low-rate subnet attempting to warehouse a slot extracts emissions much more slowly, giving the market more time to determine that it is not building anything. The emission gate further reduces what an unsupported tail subnet can extract. Expanding the total number of subnet slots makes each slot cheaper, allows more real projects to compete and makes hoarding any individual slot less valuable.
Historical simulation should still estimate the cost of maintaining a weak subnet at 25%, 12.5% or 6.25% under both linear and softer deregistration adjustments.
Owner and investor incentive differences
Owners choose the rate, while investors and subnet participants bear part of its consequences. Transparent interfaces, change delays and predictable cooldowns are therefore essential. A subnet whose owner repeatedly selects an unsuitable rate should lose market support.
Insufficient operating rewards
A resource-intensive subnet may select too little emission to operate effectively. If it cannot produce its commodity at the selected rate, it must increase the rate, attract outside revenue or improve its efficiency.
Open technical questions
This proposal defines the intended economic behaviour more precisely than the code changes needed to implement it. Technical contributors are requested to help determine:
How should the emission-gate curve be adjusted so that 50% and 25% remain viable choices for lower-ranked subnets while explicitly controlling the resulting effect on leading-subnet allocations and the tail's total accepted emissions?
What protocol treatment is required to burn declined TAO while preserving current total-issuance and halving accounting?
Should halving the selected rate always double the deregistration adjustment, or should the benefit increase more slowly at very low rates?
How long should the rate-change cooldown and gradual activation of the additional deregistration adjustment last?
What should the initial minimum rate be, and what evidence should justify enabling 12.5%, 6.25% or lower modes?
Could a low-rate subnet maintain an adjusted moving price and occupy a slot too cheaply, even though it receives fewer emissions?
What protocol changes are needed to apply the selected percentage consistently across all existing protocol-funded emissions, so that no emission route remains at 100% by mistake?
If this system had existed previously, how much TAO would have been kept out of circulation, how much alpha issuance would have been avoided, which subnets would have survived or been deregistered, and how cheaply could weak subnets have maintained slots?
These questions should be resolved through community discussion, historical modelling and testnet implementation before a final specification is submitted.
Evaluation criteria
A successful implementation should demonstrate that:
Reduced-rate subnets genuinely reduce the amount of newly emitted TAO entering circulation and reduce actual alpha issuance.
Research and early-stage subnets receive more evaluation time per unit of emission consumed.
Full-rate subnets remain able to fund expensive operations and scale proven products.
Full-rate subnets that fail to create value face greater market and deregistration pressure.
Proven lower-rate subnets can graduate smoothly towards higher rates.
Unaccepted emissions are not silently redistributed in a way that recreates the original supply pressure.
Low-rate modes do not make price manipulation or slot warehousing economically trivial.
The mechanism remains understandable to owners, investors, miners and validators.
Testing and implementation
No implementation is included at this stage.
The recommended next step is a simulation using historical moving alpha prices, subnet emission allocations and deregistration events. The simulation should compare:
Existing rules.
The linear model, where halving the selected emission rate doubles the deregistration adjustment.
Softer models, where the adjustment grows more slowly at very low rates.
Different cooldown periods and schedules for gradually activating the deregistration adjustment.
The current emission gate, softer gate curves and no emission gate, each tested with selected rates.
Different total numbers of subnet slots.
Initial rates of 100%, 50% and 25%, followed by possible 12.5% and 6.25% modes.
References
Bittensor Emissions
The V440 Emission Gate
Current Subnet Deregistration Rules