Bittensor Locks In: Conviction Ships

Conviction went live on Bittensor mainnet today. Where a subnet switches it on, every emissionEmissionsNew tokens created and distributed by a blockchain protocol over time as rewards to validators, stakers, or miners. Emissions fund network security and participation at the cost of diluting existing holders.Like a company that pays employees partly in newly printed shares. Every year the total number of shares goes up, which means existing shareholders own a slightly smaller slice of the same company unless the company grows faster than the printing.Read more → its owner earns locks the moment it lands, and the lock carries a score the network treats as a measure of skin in the game. Subnet owners who opt in can no longer build a position, capture emissions, and dump on the way out without it costing them visibly. That’s the point.

This is the governance answer the Templar exit demanded. We covered the exit itself in our Covenant deep-dive, which closed with the line “the governance score is now under active review and will be updated as the dispute resolves.” Conviction is the resolution mechanism. Whether it works is a separate question.

The mechanic, briefly

The locking rule is simple. The maths around it is less so, but the practical version is short.

Fact: Conviction = Stake × Time, and the runtime means that literally. The LockState struct on the deployed chain carries a locked_mass that decays exponentially on a roughly 90-day half-life, and a conviction score the code documents as the integral of locked mass over time. The score starts at zero, matures upward the longer stake stays locked, and only decays once the mass underneath it is gone. A perpetual, non-decaying lock mode sits alongside the decaying one.

Take: This is closer to veCRV than to anything novel. Curve gave you more weight the longer you committed, and Conviction does the same thing by integration rather than by contract length. What it removes is the fixed term: you do not sign up for four years, you accumulate weight for as long as you stay and start losing it when you leave. The signal is “are you still committed?”, and unlike veCRV it is answered continuously.

The opt-in question matters more than the curve shape. Auto-locking the owner cut is a per-subnet switch, OwnerCutAutoLockEnabled, and it defaults to false. The runtime comment is explicit that auto-locking is disabled unless a subnet turns it on, so this is a tool subnet owners adopt rather than a rule the network imposes. Owners can also lock voluntarily without the switch, and most subnets carry an owner lock of some size.

For everyone else, the change is mostly second-order. Validators and miners are not directly locked under Conviction; the binding is on subnet owners specifically. The secondary effect on alpha price action is real but bounded by adoption: subnet-owner sell pressure, historically one of the loudest signals in dTAO, is throttled on the subnets that switch the lock on and unchanged everywhere else.

Why now

On 10 April, Covenant AI sold roughly 37,000 TAO without warning and announced their exit from Bittensor in a public letter accusing the network’s founder of unilateral governance interference. TAO dropped ~15% in the hours after the announcement. SN3 alpha holders, who had bought into the Templar narrative on the back of the Covenant-72B achievement, found themselves holding an asset whose owner had repudiated the network.

The mechanical lesson was straightforward. Under the pre-Conviction model, a well-positioned subnet owner could quietly accumulate alpha through emissions, then liquidate at scale on their way out. The governance signal arrived after the financial exit, not before. If the goal of dTAO is to align subnet owners with the network’s long-term health, the absence of any binding commitment was a structural hole.

Conviction gives the network a tool for that hole. Not perfectly, and not automatically. A determined exiter can still wait out the decay window and leave with a lower-weight position, but on a subnet that has switched the lock on, the network sees them coming and the alpha holders downstream get earlier signal. The architecture is built around making exits public events rather than private trades. Whether it does that depends on how many owners adopt it, which is the thing to watch rather than the launch itself.

What changes for the three subnet camps

Three constituencies feel this differently.

Subnet owners. The cost of staying just dropped relative to the cost of leaving. Operators committed to long-term subnet performance lose nothing they were not already foregoing. Operators looking at Bittensor as a finite extraction opportunity now have a visible meter ticking against them. Expect some attrition at the margins, particularly from subnet operators whose alpha gains have outpaced their product traction.

Alpha holders. The most exposed group under the old model. Conviction is not a guarantee against an exit, but it converts what used to be a stealth liquidation into a public, scheduled, observable event. That is a real improvement in the information landscape. The bet you are making when you buy alpha now includes a Conviction-aware view of the subnet owner’s incentive to stay.

The OTF and the broader network. The governance credibility hit from Covenant was the largest the network has taken publicly. Conviction is a fast response. Whether it is the right response depends on how it interacts with the genuine concerns Covenant raised about unilateral interference, which remain unresolved on the substance. A locking mechanism makes exits costly. It does not, on its own, make the network’s emission decisions more rules-based.

What we’re watching

A few questions Conviction does not answer.

The first is whether the governance half ever switches on. Conviction’s voting weight depends on a separate per-subnet flag, VotingPowerTrackingEnabled, which also defaults to false, and the network-wide voting-power totals currently read zero. The locking machinery is live and accruing; the thing that turns a lock into a vote is not. Until that changes, Conviction is an economic commitment device wearing a governance label, and the first contested vote cannot happen at all.

The second is the interaction with the Bittensor halving cycle. Locked emissions during a halving window decay against a lower-issuance schedule, which changes the effective opportunity cost of staying versus exiting in ways we have not seen quantified yet.

The third is whether the next subnet owner who decides to leave does so under Conviction or by exploiting whatever edge cases the locking logic has. The Templar exit was unusual because it was public. Most subnet abandonments are quieter, and the network’s ability to learn from them depends on the visibility Conviction is supposed to create. The next quiet fade is the test.

What this means for the dashboard

For the Bittensor project review, the governance score that has been under active review since the Covenant exit can now be reassessed. Conviction is an improvement to a previously vulnerable layer, but an opt-in one with its voting half switched off, so it earns less credit than the launch framing suggests. The censorship-resistance dimension is unaffected. Token-distribution fairness benefits only on the subnets that adopt the lock.

For the tokenomics survey, the subnet-emission share data still tells most of the story. What changes is the read on which subnet owners are likely to still be around to capture those emissions in twelve months. Conviction is the network’s bet that the answer is “more of them than before.”

We will be tracking the first three Conviction-era subnet-owner exits as they happen. The mechanic is only as good as the behaviour it shapes, and the behaviour is only legible once someone actually leaves under it.

Correction, 15 August 2026. This entry originally described the lock as mandatory and network-wide, and the score as starting at full weight and decaying linearly under a 30-day EMA. Both came from the BIT-0011 proposal rather than the shipped runtime. Reading staking/lock.rs on the deployed chain, auto-locking is a per-subnet switch defaulting to off, the conviction score is the integral of locked mass and matures upward rather than decaying from full, and the voting-power tracking the governance claim depended on is switched off network-wide. The take about the curve being veCRV inverted was built on the wrong shape and has been rewritten. Full audit: docs/mechanism-audit-bittensor-2026-08-15.md.