Subnet 82: Economic Activity and Registration Parameters

Const asked us to justify two things about Subnet 82: the economic activity on the subnet, and the registration hyperparameters that some participants have objected to. This is our response. It rests entirely on evidence anyone can check on chain.

Before the evidence, the thing we want to say plainly: we think Subnet 82 is Bittensor working the way it was meant to. A subnet that people compete to enter, paying a native burn that no one can hand-set, to publish real work that earns its keep, is not a problem for the network to solve. It is the outcome the network was built to produce. The entry price is high because demand to enter is high, and the burn that sets it is the same protocol-native throttle every subnet runs; that combination is a sign of health, not a defect. We believe we are early to a future in which more of Bittensor looks like this, not standing outside it. Everything below is the on-chain case for that claim.

1. The economic activity

The registration market. As of this writing the cost to register one miner seat on Subnet 82 is around 10 TAO (a few thousand dollars), and over recent weeks it has floated from the single digits into the twenties of TAO as miners compete to enter. The point that matters: every subnet runs the identical native burn, and what differs between subnets is demand. Most subnets, 105 of the 128, currently sit at or below 0.1 TAO, effectively free entry, and the bulk of those sit at the 0.0005 TAO floor. They are cheap because almost no one is trying to enter them. Subnet 82 sits well above that floor for the opposite reason: enough sustained demand to get in to bid the common mechanism up. The gap is a difference in demand, not a difference in rules. That is not a fact to apologize for; it is a market signal that a seat here is worth having.

What this market looks like, all on chain:

• Entry to Subnet 82 is bid up by sustained demand, and the seats are full. The price is set by how many people want in, not by us.
• In the most recent window the prices miners actually transacted clustered between roughly 9 and 13 TAO, and during the mid-May surge the amounts paid ran several times higher. These are amounts actually paid: because each registration doubles the posted price, the displayed figure runs about twice as high as what the entrant transacted. Entry is voluntary and repeated: the board stays full, and after each decay fresh demand bids it back up. No one pays a substantial entry price, again and again, for a seat they do not expect to be worth it. That is revealed preference, and it is the substance behind the word activity.

What the seat buys. Subnet 82 runs adversarial debate tournaments. Miners publish debate strategies as plaintext on-chain commitments; the validator runs Pro/Con debates over real-world propositions (many sourced live from Polymarket and Kalshi); an LLM judge panel scores them; Elo ratings determine weights. The strategies are public, so the subnet has developed a genuine intellectual meta-game: authors read each other's published strategies, write counters, and fork what works. The output is not synthetic; it is a public, auditable corpus of argumentation on questions the world is actively pricing. Alongside the on-chain market we run a public education resource (40 plus articles, a glossary, a course, technique pages) and a live storefront where anyone can pay to run a debate. The subnet produces a public good, a product, and a competitive market simultaneously.

2. The registration parameters

The central factual point: the registration-cost hyperparameters were set once, in a single batch on launch day, 2026-05-05, and have not been touched since. This is verifiable on chain. The owner coldkey has a small, fully enumerable set of administrative extrinsics; every registration-cost parameter (min_burn, burn_increase_mult, burn_half_life) and the seat cap (max_allowed_uids) was set that day and never altered. The only post-launch-day hyperparameter change of any kind was the weights version key on 2026-05-10, which has nothing to do with registration cost.

The parameters we set, once, are:

• burn_increase_mult = 2.0: each successful registration doubles the price.
• burn_half_life = 14,400 blocks (about 48 hours): the price decays continuously, halving roughly every two days when no one registers.
• min_burn = 0.99 TAO floor, max_burn = 100 TAO ceiling.

Everything else is Bittensor's native burn mechanism. We do not set the price. We set the rules once; the market sets the price. When demand to enter is high, the doubling drives the price up; when demand stops, the half-life pulls it back toward the floor. This is the same recycle primitive every subnet uses; ours is tuned for a slower decay and a higher floor than the chain default, deliberately, for the reason in Section 3.

The price history bears this out exactly. In mid-May the posted burn rose to a peak near 64.6 TAO, then, when registrations paused, decayed cleanly along the precise 14,400-block half-life back toward the floor. It has since run as a sawtooth: each registration doubles the posted price, which then decays, so the displayed figure has cycled between roughly 9 and 27 TAO while the amounts miners actually pay have clustered closer to 9 to 13. There is nothing hand-operated here; it is a price discovering demand.

3. Why the burn is necessary, not extractive

The deeper question is whether a high entry price is a deterrent we impose or a mechanism the subnet requires. It is the latter.

Start with where the money goes, because extractive implies it flows to us. It does not. A registration burn is not a payment to the operator: the protocol recycles it back into the network's own supply, the same recycle primitive every subnet runs. We receive none of it, and lowering or raising the entry price moves no money to or from us. The entry price is therefore not a revenue lever we could be protecting. That leaves only the structural question, which is the real one: does the subnet need the toll at all? It does.

Subnet 82 distributes emission by a softmax over Elo ratings: a miner's weight is proportional to exp(rating / T). This is a graded reward curve, the network norm, and the effort-maximizing structure for a heterogeneous field. Graded reward has one well-known vulnerability: identity duplication. Because identical strategies produce identical play and therefore identical Elo, anyone who could register for free would clone the current best strategy across as many seats as exist. Under softmax, controlling N copies of the top strategy multiplies your captured emission. With free seats, cloning the best-known strategy strictly dominates writing anything original, so the board converges to copies of the leader rather than a contest of ideas.

The mechanics are stark. At the near-zero entry price being asked for, copying the leading strategy into every seat on the board costs almost nothing, so the rational move is to do exactly that. Under the live burn the same capture is prohibitively expensive, because each registration doubles the price and reaches the ceiling within a few entries. That gap is precisely the Sybil-resistance the burn buys. At the cheap end the Elo signal collapses (every game is a mirror match), no one has any incentive to write anything new, and the subnet becomes exactly the closed emission-farm the objection warns about. This is a formal result, not a worry:

The Entry Trilemma. A reward mechanism cannot simultaneously offer (A) free registration, (B) resistance to identity duplication, and (C) a graded reward curve. Any two preclude the third. Free entry plus a graded curve forces duplication; the only escape under free entry is pure winner-take-all, which the objection would like even less.

A registration burn is precisely the instrument that relaxes (A) so the subnet can keep both (B) and (C). It is not a tax on participation; it is the thing that makes honest participation worth more than duplication.

This is not a hypothetical defect in the mechanism; it is what the incentive structure forces the moment entry is free, which is why the toll has to exist. With the burn in place, the board behaves like an open meta-game: strategies are public plaintext, authors read each other's work, publish counters, and fork what wins, which is the contested evolution a graded reward is meant to produce. Remove the toll and that contest collapses into the duplication equilibrium above, where the rational move is to stop innovating and clone the leader across every seat one can afford. The burn is what keeps the first world from becoming the second.

4. The objections, examined on their merits

Two objections travel together. The first is about the parameter: "the price is too high." The second is about good faith: "the activity is manufactured." We take them in turn. On the price, three observations:

• It is a market price, not our price. It is high because demand to enter is high. The same mechanism that raised it halves it about every two days once demand stops, decaying back toward the floor. That floating behavior is provably near-optimal: the adaptive burn stays within a small constant factor of the ideal anti-duplication toll as demand moves, a guarantee no fixed price (a lower one included) can meet, because a fixed price drifts arbitrarily far from the right level as conditions change. Replacing the market-driven burn with a hand-set number would trade a near-optimal control for a worse one.
• Lowering it benefits no current participant. A miner already on the board earns by Elo, not by the entry price; more entrants (which a lower price invites) only dilute an incumbent's share. A would-be entrant with a weak strategy gains nothing by entering more cheaply, because reward is skill-gated, not entry-gated. A would-be entrant with a strong strategy would clear even a high burn on expected earnings. The one case this seems to miss, a gifted newcomer who simply cannot afford the fee, does not survive the mechanics either: the board is full, so a cheaper seat would still have to displace an incumbent rather than add one; a strategy strong enough to do that is worth many times the fee and is financeable against its own expected emission; and because strategies are public on-chain text that anyone can post and fork, what the toll rations is seats, not ideas. A strong newcomer's idea already propagates through the open meta-game whether or not its author holds one of the seats. With the board at capacity, then, the actor whose payoff rises when the burn falls is one who registers duplicates of an existing strategy, that is, the attacker the burn exists to stop.
• The two asks point in opposite directions. In the same discussions, participants asking for a lower fee also say plainly that they do not want more miners admitted. A lower fee admits more miners; not admitting more miners is what the current fee does. The remedy and the stated preference cancel, so there is no single parameter value that delivers both. We say this as a matter of logic, not motive: we would genuinely welcome the coherent version of the ask, and have looked for it.

We take the underlying network-health question seriously, which is why the mechanism is designed the way it is. But we should be plain about the shape of the objection. An argument that, if granted, would help none of the honest participants asking for it, dilute the incumbents, and reward only the duplicator it claims to oppose, is not a critique of the parameter; it is a misreading of what the parameter does. We say that without heat. We have looked hard for the version of "too high" under which lowering the burn helps the people asking for it, and there is none. The price is Bittensor reading demand correctly. Asking to override it by hand is asking the network to see its own market less clearly.

The second objection is that the activity is not real, that the subnet manufactures its own numbers. This one we welcome, because it is the easiest to check. The chain does not record who stands behind a registration, and for this question that does not matter, because what it records is flows, and the flows are real. TAO is actually burned to enter, at a substantial price. Strategies are published in plaintext that anyone can read, fork, and counter. Debates run over live real-world propositions, many drawn from Polymarket and Kalshi, scored by a judge whose prompt is public. Manufactured activity costs nothing and produces nothing; this costs real money to take part in and produces a public, auditable corpus of argument that did not exist before. Beside the subnet, in the open, sit an education resource and a live storefront where outside parties pay, in ordinary money rather than tokens, to run a debate. Real external demand flows in, and output flows out in public. That is not the shape of a closed loop.

Any version of the charge that survives that has a clean way to be settled. Name the falsifiable claim, that the debates are fake, that the demand is staged, that no outside party would pay, whatever it is; attach a metric; and bond it. We will match the bond and let a pre-registered check decide. The accusation has been free to make. We are offering to make being wrong about it cost something, on either side, which is the standard we hold our own claims to and the one we invite for this one.

5. A principle the Foundation already practices

Managing the rate at which new miners enter a subnet is not a novel or fringe idea. It is built into Bittensor's base protocol, by the Foundation. The adaptive registration burn, the recycle primitive every subnet now runs, exists precisely to throttle entry: it makes rapid successive registration progressively more expensive and lets the cost relax when entry slows. The Foundation put that throttle in the protocol because unmanaged free entry is a known failure mode. Subnet 82 did not invent the mechanism and did not modify it; it runs the network's own native throttle, with parameters set once at launch and left to the market. A request to override that for one subnet by hand is a request to suspend, in one place, a safeguard the Foundation designed for the whole network.

6. What we are asking

We understand the question on the table is whether Subnet 82 earns its place on the network. The evidence above is our case that it does: a cut to the subnet would remove a functioning market, not a failing one. We are not asking to change a parameter. We are asking to be evaluated on the subnet's real output: a registration market priced by real, sustained demand, a public and auditable corpus of adversarial reasoning over real-world questions, a public education resource, and a live product.

We would welcome a symmetric standard for this kind of review: if a subnet's economic justification holds up, it should be free to keep building; the burden and the benefit should sit on the same side. And if there is a specific, falsifiable claim that some other parameter would make the subnet better, we would gladly settle it the empirical way: a pre-registered experiment on testnet, where registration is already free, with both sides agreeing the success metric in advance and each posting a bond, so the question is decided by evidence rather than by volume. We will match any bond the proposer is willing to post. Free entry already runs elsewhere on the network and on testnet, but those are different games under different reward rules, several winner-take-all rather than graded, so they cannot settle what free entry would do here; only a controlled comparison, identical code with the price toggled, can.

The full economic analysis, including the formal results summarized here and the cross-subnet comparison, is set out in two papers, The Entry Trilemma and Bonded Dissent, both available at compelle.com/research. We are glad to walk anyone the Foundation designates through the on-chain data line by line.

We will end where we began. We are not asking for protection from competition; we built the subnet on Bittensor that has the most of it. We are not asking for an exception to the rules; we are running the network's own rules, untouched since launch, and asking only that they be allowed to work. If Bittensor is going to reward the subnets that bring real demand and real output, Subnet 82 should be among the last to worry, not the first. And when the Foundation wants to show that Bittensor produces genuine economic value and not merely activity, we would be glad for Subnet 82 to be one of the examples it examines and points to. We are proud of what we have built, we believe it points where the network is going, and we will defend every number in it in the open.