Selfish mining

Withholding a solved block is normally self defeating, since a rival can publish first and your work becomes worthless. A 2013 paper by Ittay Eyal and Emin Gun Sirer showed that above roughly one third of network hash power the arithmetic flips, and with a propagation advantage the threshold falls toward 25 percent. Nothing resembling it has been documented on bitcoin's main chain.

How it works

Selfish mining converts a private lead into wasted electricity for everybody else.

A selfish miner finds a block and says nothing. From there the play branches. If the rest of the network publishes a block while the secret one is still hidden, the selfish miner releases immediately and forces a race between two tips of equal height, which it can win if enough of the honest hash rate happens to hear its block first. If instead it finds a second block privately, it now holds a two block lead, and it can let honest miners keep grinding on a branch that is already doomed. When they finally publish, the hidden pair is released, their block is discarded, and their work earned nothing.

The payoff is relative, not absolute. Every discarded honest block reduces the total number of blocks the network accepts, so in the short run the selfish miner earns less bitcoin than it would have earned by playing straight. What changes is the ratio: it captures a larger slice of a smaller pie. The pie is then restored by the difficulty adjustment, which recalculates every 2,016 blocks and lowers the target because blocks arrived slowly. Only after that retarget, up to two weeks later, does the strategy pay. Anyone attempting it therefore has to fund a fortnight of reduced revenue while behaving in a way that is statistically visible from outside.

The threshold depends on network topology, which is the part of the paper most summaries drop. The relevant variable is what fraction of honest miners build on the selfish block during a race. If that fraction is zero, selfish mining needs more than a third of total hash power to pay. If it is one half, a quarter is enough. If a miner could reliably reach every peer first, the profitable share would fall much further, which is why block propagation speed is a security property rather than a performance detail.

Where you see it

Selfish mining shows up in academic literature and in the design of relay software, not in observed bitcoin history.

Three practical facts keep it there. Propagation is close to instant: compact block relay, added in Bitcoin Core 0.13 in 2016, sends a short sketch of a block that peers reconstruct from their own mempools, and dedicated relay networks cut the rest, so the window in which a withheld block wins a race is tiny. Pools are also public. Almost every block carries a pool tag in its coinbase script, and a pool producing an odd pattern of last second releases would be identified quickly by hash rate it does not own, since its customers can point their machines elsewhere within minutes.

The term also gets confused with two neighbours. Block withholding is sabotage aimed at a pool, where a participant submits partial shares but discards the one that actually solves a block, costing the pool rather than the network. A 51% attack rewrites confirmed history. Selfish mining does neither: it breaks no consensus rule, reverses no payment, and steals revenue from other miners rather than coins from users. For anyone buying or holding, the only symptom would be a higher rate of stale blocks, and confirmations would still mean exactly what they mean today.