Mining pools, shares and the centralisation question

A share is a proof of work that is much easier to produce than a block. The pool sets a share target far above the network target, so that a participant finds a share every few seconds rather than every few years. A share is not a valid block and earns nothing from the network; it is a sample that demonstrates the participant was searching, and its difficulty is a measure of how much work went into finding it.

Because shares are found at random intervals, the count a participant submits over a short window is a noisy estimate of their true share of the pool's hash rate. Over a long window the noise averages out and the counts converge on the underlying contribution. This is why pools credit work over a rolling window rather than instantaneously, and why a participant who joins and leaves quickly may be credited slightly above or below their true contribution.

The share mechanism is also the pool's exposure. A participant who submits shares without ever finding a block still expects to be paid, because the pool is buying the work rather than the outcome. If the pool pays out more than it earns over a period, it absorbs the difference; if it earns more, it retains the surplus under some schemes and distributes it under others. The choice of scheme is therefore a transfer of variance between the operator and the participants, and it is the main thing that distinguishes the common payout methods.

Pay-per-share pays a fixed amount for every share submitted, regardless of whether the pool finds a block. The participant's variance is essentially zero, and the pool carries all of it. In exchange, the pool charges a higher fee and must hold a reserve to cover periods when it pays out more than it earns. It is the most predictable arrangement for a small miner and the most demanding for the operator.

Proportional payment distributes each block's reward across the shares submitted during the round that produced it, in proportion to their difficulty. The participant's income is exactly the pool's income, shared out, so the operator carries no variance and charges a lower fee. The participant carries the variance instead, including the risk that a round is unusually long and their effective rate falls.

Pay-per-last-N-shares is a middle position and the most widely used in practice. Each block's reward is distributed across the last N shares submitted, where N is chosen to be roughly the number of shares a round is expected to take. Because the window slides, a participant is paid for work that spans more than one round, which smooths the variance considerably without requiring the operator to hold a large reserve. Variants such as pay-per-share-plus and full-pay-per-share adjust the same trade-off in different directions.

The schemes differ in who bears the variance and how much the operator charges for bearing it. They do not differ in expected return before fees, because the underlying work and the underlying reward are the same in every case. A miner choosing between them is choosing an income profile, not a better or worse deal.

Pooling concentrates a decision that the protocol leaves to miners: which transactions go into a block, and in what order. The pool operator builds the template, so the operator's policy — not the individual participant's — determines what the block contains. A pool that declines to include certain transactions effectively applies that policy to every machine pointed at it, however many independent operators those machines belong to.

The concentration is real but it is not the same as a concentration of hash rate ownership. Participants can and do switch pools, and switching is a configuration change rather than a capital expenditure. That mobility is what disciplines pool policy: a pool that censors transactions or behaves badly risks losing its participants to a competitor, and the loss is immediate rather than gradual. The threat model is therefore one of temporary coordination rather than permanent control.

The countervailing development is the growth of stratum protocols that let participants construct their own block templates while still receiving pooled payouts. Under these arrangements the pool coordinates the accounting but does not choose the transactions, so the policy concentration is reduced without giving up the variance reduction that made pooling attractive in the first place. Adoption is uneven, and the extent to which it has displaced the older model is a matter of ongoing measurement rather than settled fact.

What the protocol does not do is cap any pool's share of hash rate. There is no consensus rule that limits concentration, and no mechanism that would enforce one. The defence is economic and social rather than cryptographic, which is why the distribution of hash rate across pools is watched closely by people who care about the network's resistance to coordinated action. The hashrate page explains the quantity being distributed, and the proof of work page explains why the work itself is the security mechanism. The miner revenue page covers what the pool is actually distributing.