The Network Nervous System

A proposal is a typed action with parameters. The documentation groups proposals by topic — among them subnet management, node management, network economics, participant management, and the governance configuration itself — and each topic has its own voting rules. A proposal is decided by the votes of neurons, weighted by the amount of ICP staked in them, and it passes when the votes in favour meet the topic's threshold.

Voting is not a one-off act. A neuron can be configured to follow other neurons on a topic, so that its votes are cast automatically in line with a chosen delegate. The documentation describes this as the mechanism that makes governance tractable: a holder who does not want to evaluate every subnet-management proposal can follow a neuron that does, while retaining the ability to vote directly on anything they care about. Following is per topic, so a neuron can delegate economics and vote directly on protocol changes.

The documentation also describes the reward mechanism. Neurons that vote earn voting rewards, and the reward depends on the neuron's stake and on its dissolve delay. A neuron that is dissolving has a shorter delay and earns less; a neuron locked for a long period earns more. The intent is to align the people who decide the network's future with its long-term health rather than with the next proposal.

A neuron is a staking position. It holds ICP, it has a dissolve delay that determines how long the stake is committed, and it has an age that grows while the neuron is not dissolving. The documentation describes the voting power of a neuron as a function of its stake and its dissolve delay, with a bonus for age. A neuron that is dissolving loses voting power as its delay runs down, which is what makes the commitment meaningful.

The dissolve delay is the central economic parameter. A neuron with a long delay has more voting power and earns more rewards, but its stake cannot be withdrawn until the delay elapses. A neuron can increase its delay at any time; decreasing it starts a countdown that can be stopped before it completes. The documentation presents this as a deliberate trade of liquidity for influence.

The documentation is also explicit that staking is not risk-free. The value of the staked ICP is exposed to the market, and the rewards are paid in newly minted ICP and in a share of the transaction fees the network collects. A neuron's return therefore depends on the network's activity and on the price of the asset, not only on the staking parameters.

The NNS controls the network's infrastructure. It creates and configures subnets, it adds and removes node machines, it manages the system canisters — the registry, the governance canister, the ledger, the cycles minting canister, the Bitcoin canister and the others — and it sets the network's economic parameters. It also controls the version of the protocol that the network runs, which is how upgrades are deployed.

The distinction between the NNS and an SNS is worth keeping clear. The NNS governs the network; an SNS governs a single application that has chosen to hand its control to a community. The SNS DAOs page covers the application-level case, including how a project launches one and what the swap does.

The documentation notes that the NNS is itself a set of canisters running on the network, which means it is subject to the same properties as anything else: it is upgraded through proposals, its state is replicated, and its behaviour is verifiable. That self-referential structure is what allows the network to be governed without an external authority holding the keys.