The difficulty adjustment: Bitcoin’s two-week thermostat, explained

Difficulty is the size of the target miners are aiming at

Mining is a guessing game. A miner assembles a block, runs its header through a hash function, and checks whether the result is at or below a number called the target. Almost every try fails, so the miner changes a field called the nonce and tries again, billions of times a second. The Bitcoin Wiki defines difficulty as a measure of how hard it is to find a hash at or below a given target: the smaller the target, the harder the game. Difficulty is expressed relative to the easiest target the network allows, so block 0, mined at that easiest setting, has a difficulty of exactly 1.

Satoshi expected hardware to get faster and interest to come and go. The white paper says the proof-of-work difficulty is determined by a moving average targeting an average number of blocks per hour, and that if blocks are generated too fast, the difficulty increases. Without that rule, every improvement in mining hardware would make blocks arrive faster, the supply schedule would run ahead of plan, and the network would have less time to spread each block around before the next one appeared. Our explainer on proof of work covers the guessing game itself.

Every 2,016 blocks, the network checks its watch

The numbers come from two settings in Bitcoin Core’s chain parameters. The target spacing is 10 minutes, and the target timespan is 14 days. Two weeks at one block every ten minutes is 2,016 blocks, and that is the adjustment interval. The code in pow.cpp checks whether the next block’s height is a multiple of that interval; if it is not, the target stays exactly what it was.

At the boundary, the node reads the timestamps in the first and last block headers of the period just ended and subtracts one from the other. The developer documentation states the ideal value: 1,209,600 seconds. The new target is the old target multiplied by the actual timespan and divided by the ideal one. If the 2,016 blocks came in a week, the target halves and mining gets twice as hard. If they took four weeks, the target doubles and mining gets twice as easy. Every node does this arithmetic independently and gets the same answer, because it uses only data that is already in the chain.

There is a small, permanent quirk. The period is 2,016 blocks long, but the timestamps the code compares are those of the first and last blocks in the period, which are 2,015 intervals apart. The developer documentation calls it an off-by-one error in the original implementation that creates a slight skew. It cannot be fixed without changing a rule every node enforces, so it stays.

The 4x clamp keeps one wild fortnight from breaking the chain

Before the division, pow.cpp clamps the measured timespan. If it is less than a quarter of the ideal, it is set to a quarter; if it is more than four times the ideal, it is set to four times. In plain terms, difficulty can go up by at most 300% or down by at most 75% in a single step, and the developer documentation states the same limits in those words.

Why bother? The clamp limits how far any single measurement, honest or otherwise, can move the target. It also means a sudden collapse in mining power cannot be fixed in one go. If most miners vanished at once, blocks would slow to a crawl, and the next adjustment could only ease the target by a factor of four, so the recovery would take several periods, each of which would take longer than two weeks to complete. The clamp trades speed of recovery for protection against a wild swing.

There is one more guard. The target can never be set easier than a fixed ceiling called the proof-of-work limit, so difficulty never drops below 1 no matter how few miners remain. The main network has no exception to this; the chain parameters switch off a testing-network rule that allows minimum-difficulty blocks when the chain stalls.

The target adjusts after observed block timing

Difficulty is adjusted at scheduled block-height boundaries using the chain’s timing data. It reacts to past conditions rather than measuring every miner’s equipment directly.

If mining power changes sharply between adjustments, block intervals can differ from the ten-minute target. Even with unchanged power and difficulty, individual intervals vary. “About two weeks” describes the intended adjustment timescale, not an exact calendar schedule.

The biggest drop on record came after China’s mining ban

The clearest demonstration came in 2021. China, then home to a large share of the world’s mining, ordered mines shut, and CoinDesk reported that the crackdown cut the network’s total hash power by more than half from its record earlier that year. Blocks slowed. The adjustment on 3 July 2021 cut difficulty by 28%, reported at the time as the largest downward move in Bitcoin’s history, and CoinDesk noted that miners still running could expect their profitability to double as a result.

Nothing about the event required a decision from anyone. The rule did what the white paper described: measure, scale, wait, repeat. The story of the ban itself is in our article on China’s 2021 mining ban, and Satoshi’s early notes on difficulty and hardware are collected in the proof-of-work and difficulty correspondence hub. For how the adjustment fits with the rest of block building, see the how the blockchain works dossier.