Selfish Mining is a strategy of withholding valid blocks and timing their release. Its aim is to earn a larger share of accepted blocks than the miner’s share of computing power. It neither allows bitcoins to be created outside the rules nor guarantees profit: connections, other miners’ behaviour and costs matter too.
Peter runs a Mining Pool, a group of miners that combine computing power and share rewards. A block groups transactions into a batch added to the shared history. Peter’s group finds a valid block. Normally it would announce the block so others could build on it. This time it keeps the block private and starts looking for its successor. Anna and the other miners continue working on the old shared block. They do not know Peter has found another way to extend the history. This starts a private branch, not a new currency. [1] [2]
Peter might find the next block, but Anna might find one first. Mining machines repeat calculations with modified block data until they find a numerical result that meets the required limit. Mining is not a regular race in which the faster machine always finishes first; each discovery is random. The strategy responds to both branches, sometimes continuing in secret, sometimes releasing blocks or abandoning its branch. Hashrate means the rate of computational attempts. A larger share improves the chance of finding a block without guaranteeing the next one. [1] [4]
Suppose Peter has hidden one block and Anna publishes her own block on the same predecessor. Peter reveals his, creating a tie. Different miners may hear about different blocks first and start extending different branches. What happens next, and how messages spread, helps decide the winner. The deciding measure is accumulated work in a valid branch, not the number of people supporting it; at equal difficulty, more blocks mean more work. Difficulty expresses how hard that result is to find on average. Higher difficulty means more attempts on average, not a fixed time for each block. [1] [3] [4]
If Peter’s branch eventually wins, Anna’s competing block may remain outside the accepted history as a Stale Block. It was valid, but its reward cannot be spent on the accepted chain. Peter wants a greater proportion of other miners’ work to be wasted than of his own. Transactions from the abandoned block need not disappear forever: if still valid and compatible with the accepted history, they can be included later. [1] [4]
Ittay Eyal and Emin Gün Sirer described the well-known strategy in a 2013 paper. They compared a miner’s share of accepted blocks with its share of computing power. That differs from hourly earnings after electricity bills. Withholding can slow the public chain’s growth, and the subsequent difficulty adjustment changes the economics. Cyril Grunspan and Ricardo Pérez-Marco later emphasised time and that adjustment. A higher relative reward share is therefore no promise of immediate net profit. [1] [7]
In the original model, α is the attacker’s share of computing power. γ is the share of the other miners’ power that joins its branch in a tie. With γ = 0, the relative-advantage threshold is 1/3; with γ = 1/2, it is 1/4. The model predicts an advantage above the respective threshold. These are not universal safety boundaries for the real network. Other strategies, delays and assumptions change the result. The essential point is that both computing power and which branch others extend matter. [1] [6]
A quickly delivered block has a better chance of attracting miners in a tie. A network advantage can therefore complement computing power. A group offering consistently better returns might attract more miners and increase concentration. That describes an incentive, not proof that every large pool attacks or that others inevitably join. Nor can a model parameter be treated as a measured property of a particular operator without evidence. [1] [6] [8]
Anna may pull ahead before Peter’s strategy pays off. Then Peter receives no reward despite spending energy and operating machines. A threshold calculated for a long-term average does not promise victory in any individual round. A few successful rounds also cannot establish profitability after costs and changing conditions. A withheld block is a risky opportunity, not a reward safely stored in a drawer. [1] [7]
A Full Node, a program that independently checks Bitcoin’s rules, still rejects invalid transactions and unauthorised coin creation. Peter must supply real Proof of Work, demonstrating the required computational effort, and follow the same rules as Anna. The strategy competes over which valid history prevails. It does not give him anyone else’s private keys, the secret numerical values used to sign transactions, or permission to spend their coins. A mining advantage must therefore be distinguished from breaking ownership rules. [2] [5]
Faster block relay, for example through Compact Blocks, which use already known transactions to transmit a block efficiently, reduces delays and some network inequalities. It does not eliminate Selfish Mining. Research also examines tie handling and economic incentives. A stale block alone proves no attack, because two honest groups can succeed almost simultaneously. Assessing suspicions needs wider context. The underlying question remains: do the rules reward prompt sharing under these conditions, or can withholding pay? [6] [8] [9]
For the clearest picture, read this entry together with Block Propagation, Compact Blocks, Nakamoto Consensus, Sybil Attack, Bitcoin, Mining. The reverse links also lead from Byzantine Generals Problem, Sybil Attack, Block Propagation, Block Withholding Attack.