Experiment Lab
Change one thing, watch the whole system respond
Mining power, block discovery, difficulty and the mempool on one machine, driven by one simulation. Choose an experiment, predict what will happen, then run it.
Explanation level
What happens when half the network's mining power disappears?
Half the miners switch off. Blocks keep coming, but about twice as slowly, because the puzzle has not changed. Only after 2016 blocks does the network look back, see that they took too long, and make the puzzle easier.
Watch the clock: its hand (days) runs ahead of the orange ring (blocks). Press “Run to the next adjustment” and compare before and after. Then try removing even more power yourself.
The hash rate halves at block 504. The expected interval doubles to about 1,200 s; in this seeded run the period's 2015 intervals take about 23.4 days instead of 14 (on average about 24.5 days), so the retarget multiplies difficulty by 1,209,600 / actual timespan (here about ×0.6, on average about ×0.57, not ×0.5, because a quarter of the period ran at full speed). The next period corrects the rest.
What happens when twice as much mining power joins?
Twice as many miners join. Blocks come about twice as fast until the end of the period. Then the network notices the period was too short and makes the puzzle harder.
Watch the orange ring race ahead of the clock's hand, and the governor slide toward higher difficulty before anything changes on the dial.
With twice the hash rate the expected interval is about 300 s. In this seeded run the period ends after about 8.5 days (on average about 8.75 days), so difficulty rises by about ×1.65 (on average about ×1.60); the ×4 clamp is not reached.
What happens when far more people want to transact?
Mining stays the same, but far more people want to pay. Blocks can only carry so much, so a queue grows and paying a higher fee becomes the way to get in sooner.
Watch the flow meter: the orange (arriving) bar is longer than the grey (leaving) bar. The columns fill, and fewer of them light up for the next block.
Demand rises to 1.6 × 6 tx/s at 250 vB, about 2,400 vB/s against about 1,667 vB/s of block space. The backlog grows until the pool limit evicts the lowest fee rates; difficulty is untouched because mining did not change.
What happens when miners leave just as demand rises?
Miners leave just as more people want to pay. Fewer blocks carry away less, while more arrives: the queue grows fast, until the difficulty adjustment brings blocks back up to speed.
Compare the flow meter before and after the change, then run to the adjustment and see how much of the backlog the faster blocks can work off.
Hash rate ×0.6 and demand ×1.3 at block 504: block capacity falls to about 1,000 vB/s while arrivals rise to about 1,950 vB/s. The retarget restores capacity, but demand still exceeds it in this experiment, so the backlog stays at the pool limit.
Predict first (optional)
Choose your prediction, then run the experiment. The answer comes from the simulation itself, not from a button.
Run the experiment to see the answer.
Run the experiment to see the answer.
Run the experiment to see the answer.
Run the experiment to see the answer.
Simulated time has not started. The machine runs the experiment from its fixed seed.
The experiment machine loads when it scrolls into view.
Open the machine on its ownOpen the machine in its own window · Works offline; nothing is sent anywhere.
Bitcoin's rules in this experiment
- Difficulty Every 2016 blocks the target is recomputed: new = old × actual timespan / 1,209,600 s, the timespan limited to between a quarter and four times two weeks, and the result rounded to compact precision (Bitcoin Core CalculateNextWorkRequired).
- Difficulty The timespan is measured from the period's first to its last block, 2015 intervals, a long-known quirk that is now part of consensus.
- Difficulty Difficulty never reacts to a change in hash rate before the period ends.
- Blocks A block may weigh at most 4,000,000 weight units, which is 1,000,000 vB.
Assumptions of this experiment
- Blocks Block discovery is modelled as a Poisson process: each wait is exponential with mean = expected work per block / hash rate, drawn from a seeded generator.
- Time Block timestamps equal the moment of discovery, rounded to whole seconds. Real miners' clocks vary within the consensus limits.
- Mining Hash rate changes in one step at the stated block; in reality miners leave or join over hours or days. Starting hash rate: 600 EH/s, a round value of the right order of magnitude (somewhat below public estimates for 2025–2026), not live data. Block times and retargets in the experiments do not depend on its exact value.
- Difficulty The experiment starts at the beginning of a period with a target that matches the starting hash rate exactly (ten minutes per block on average).
- Mempool Transactions arrive as a seeded Poisson process at the stated rate (6 per second at the start), each 250 vB, with a fixed mix of fee rates from 1 to 200 sat/vB. Real users adjust their fees to the market.
- Mempool Miners fill each block with up to 1,000,000 vB, highest fee rate first. Real miners select transaction packages (ancestors, clusters) and some leave room unused.
- Mempool The mempool holds at most 100 million vB, roughly what Bitcoin Core's default 300 MB memory limit keeps (a node policy, not a consensus rule). Above it the lowest fee rates are evicted first. There is no expiry and no replacement in the experiment, and demand does not react to fees, so backlogs grow larger than real users would let them.