Collusion
Collusion is when participants coordinate to bend an outcome in their favor instead of acting honestly. In a Bittensor subnet, for example netuid 1, the concern is colluding validators that assign inflated weights to particular miners, or to each other, so that consensus rewards those targets beyond what their actual work merits.
References: Understanding Incentive Mechanisms
What It Would Target
The aim of collusion is to capture more emissions than honest evaluation would allow. A group of validators acting together might try to push the consensus result toward favored miners, steering a larger share of rewards their way. The attack depends on the colluders being able to move the consensus outcome by coordinating their weights.
References: Understanding Incentive Mechanisms
How Yuma Consensus Resists It
Yuma Consensus is built to blunt exactly this. The documentation describes consensus as a stake-weighted agreement point, with weights that outlie that agreement being clipped. Weights pushed beyond the consensus benchmark are not rewarded, so an inflated rating that departs from the broader agreement is trimmed away rather than paid out. The clipping is meant to punish patterns that look like collusion to favor certain miners.
References: Understanding Incentive Mechanisms, Yuma Consensus
Stake Behind Influence
Coordination alone is not enough, because influence over consensus is weighted by stake. The agreement point that defines what gets clipped is itself stake-weighted, so a colluding group can only move it if it holds enough stake to shift the weighted middle. A coalition of low-stake validators cannot drag the consensus toward its targets, which raises the cost of mounting a believable collusion.
References: Yuma Consensus, Understanding Incentive Mechanisms
Development Stage Context
The Introduction to Bittensor describes subnet development as moving from localnet to testnet and then mainnet. For collusion, that sequence changes how readers should interpret stake-weighted consensus and validator agreement examples.
In localnet, consensus resistance mechanics can be tested in an isolated environment. Localnet stake distributions do not represent production validator influence.
On testnet, validator weight and stake patterns can be exercised in a shared non-production network. Testnet consensus outcomes are separate from mainnet subnet state.
On mainnet, collusion resistance concerns live production validator stake and Yuma Consensus on subnets such as netuid 1. Observed stake distributions and consensus values depend on the selected subnet’s current chain state (Yuma Consensus).
The Bittensor Networks reference separates mainnet, testnet, and localnet. A collusion-resistance example from one environment should not be read as representing production consensus behavior in another environment.
Relationship to Yuma Consensus
Collusion and Yuma Consensus describe related parts of Bittensor’s incentive system. Yuma Consensus is the on-chain process that aggregates validator weight signals within a subnet into miner incentives and validator dividends, applying consensus clipping, bonding, and emission calculation (Yuma Consensus).
For readers, collusion names a specific part of that incentive picture, while Yuma Consensus names the consensus process that turns validator weights into the resulting incentives and dividends.
Reader Boundary
This page defines the concept and Bittensor’s consensus-level resistance at a high level. It does not claim collusion is impossible, quantify how much stake would be required, or describe how to attempt it. The stake distribution and consensus values that determine how strong the resistance is on any subnet are live chain state and change over time.
References: Yuma Consensus
Weight Copying Can Raise Agreement Without Coordination
The Weight Copying Problem describes validators reusing visible weight signals instead of independently evaluating miners. Copied weights can make separate validators look aligned even when no group agreed to coordinate.
Collusion instead names validators acting together to push inflated ratings toward chosen miners. Similar-looking weight patterns therefore do not, by themselves, prove collusion; the distinction is whether the alignment came from coordinated inflation or from copied signals (Glossary: Weight Copying).
References: Weight Copying Problem, Glossary: Validator Weights
Sybil Attacks Target UID Count, Not Coordinated Weights
A Sybil attack tries to gain leverage by registering many separate identities that appear independent. Collusion instead tries to move the consensus outcome by coordinating validator weights toward favored miners.
The two threats should be kept separate in safety reading. Multiplying UIDs is an identity-flooding pattern; collusion is a coordinated scoring pattern inside an existing validator set (Understanding Incentive Mechanisms).
References: Understanding Incentive Mechanisms, Yuma Consensus
Clipped Weights Lose Reward at Tempo Boundaries
Understanding Incentive Mechanisms states that Yuma Consensus calculates emissions within each incentive mechanism and finalizes those results at the end of each tempo. Weights that exceed the stake-weighted consensus benchmark can be clipped before that distribution step (Yuma Consensus).
Collusion resistance therefore lands at payout time, not only at the moment weights are submitted. An inflated coordinated rating must still survive clipping and tempo-bound distribution to produce extra reward (Emission).
References: Emission, Glossary: Tempo
Relationship to Weight Copying
Collusion and weight copying are related but different safety concerns in Bittensor consensus vocabulary. Weight copying names validators reusing visible weight signals without independent evaluation, while collusion names coordinated validators trying to push consensus toward favored miners beyond honest evaluation. The Weight Copying Problem describes copying risk from visible weights, and Understanding Incentive Mechanisms describes consensus clipping that punishes inflated ratings departing from agreement.
For readers, weight copying names an imitation risk from visible signals, while collusion names coordinated inflation of ratings toward chosen targets.
References: Weight Copying Problem, Understanding Incentive Mechanisms
Relationship to Validator Weights
Collusion and validator weights are related but different parts of Bittensor consensus vocabulary. Validator weights name the evaluation signals validators submit after measuring miner work, while collusion names coordinated attempts to push those signals beyond what honest evaluation would support. The Glossary: Validator Weights defines validator weights as evaluation inputs, and Yuma Consensus describes how outlying weights can be clipped relative to a stake-weighted agreement point.
For readers, validator weights name the submitted evaluation signals, while collusion names coordinated distortion of those signals toward favored miners.
References: Glossary: Validator Weights, Yuma Consensus
Relationship to Validator Trust
Collusion and validator trust are related but different parts of Bittensor subnet vocabulary. Validator trust names a subnet-local measure of how consistently a validator’s weights align with consensus over time, while collusion names coordinated attempts to bend consensus toward favored targets. The Glossary: Validator Trust describes validator trust as a consensus-alignment measure, and Understanding Incentive Mechanisms describes consensus mechanisms designed to resist coordinated inflation.
For readers, validator trust names sustained alignment with consensus, while collusion names coordinated departure from honest evaluation aimed at favored miners.
References: Glossary: Validator Trust, Understanding Incentive Mechanisms
Relationship to Netuid
Collusion and a netuid are related but different parts of Bittensor vocabulary. A netuid identifies a subnet on the network, while collusion names a coordinated consensus-risk pattern that would be discussed inside one subnet’s validator set and miner evaluation context. The Glossary: Netuid places netuid at the subnet level, and Understanding Incentive Mechanisms frames subnet-specific validator evaluation before consensus aggregation.
For readers, a netuid identifies which subnet market is in scope, while collusion names a coordinated manipulation concern inside that subnet’s consensus context.
References: Glossary: Netuid, Understanding Incentive Mechanisms