Waiting longer does not bring the next block closer. The block is not standing behind you in a line, and it has no idea how long you have been waiting.
There is a countdown on almost every block explorer: next block in 04:12. It ticks down while you watch. When it reaches zero, one of two things happens — a block appears, or the number jumps back up to nine minutes and change and starts over.
I once left that tab open for an hour on a slow Sunday, mostly out of stubbornness. In that hour I learned the only thing about Bitcoin block timing that actually matters, and it was not about fees, hashrate, or the mempool. It was that the countdown is a decoration. Bitcoin does not schedule blocks. It schedules the difficulty that makes blocks average out to ten minutes. Those are not the same statement, and the gap between them explains most of the confusion people have about confirmation times.

The First Analogy: A Geiger Counter With No Memory
A Geiger counter next to a lump of uranium clicks at a steady average — say once a second. But it does not click on the second. The gaps between clicks are random, and the distribution of those gaps has a specific and slightly insulting property: it is memoryless.
If you have waited five seconds in silence, your expected wait for the next click is still one second. It is not shorter because the counter “owes” you a click. Nothing accumulated. The counter has no state that remembers the silence.
Bitcoin’s proof-of-work is the same kind of process. Every hash attempt by the entire network is an independent lottery ticket, and the network tries an enormous number of them per second. Nobody is “close” to finding a block, because there is no such thing as close — the winning hash is a fresh random draw every time. Miners are not making progress toward a block. They are re-rolling dice at a rate measured in quintillions of rolls per second.
That is why the countdown is a lie with good intentions. It is estimating a mean, and drawing it as if it were a deadline.
The Second Analogy: The Bakery With No Timetable
My local bakery puts out a fresh tray “about every ten minutes”. Over a day that is accurate — forty-odd trays across the morning. But walk in at a random moment and your expected wait is ten minutes, whether the last tray came out thirty seconds ago or eight minutes ago.
The bakery, in other words, is not running a queue. It is running a rate. A queue tells you your position; a rate tells you only the average spacing. If you arrive and the tray is empty, the information “one just came out” is worth nothing to you, because the baker is not working through a backlog of trays. He is just working.
This is the analogy to keep, because it also shows where the bakery analogy breaks — and the break is the important part.
Where it breaks: the tray is not the block, it is the block space. Bitcoin does have a queue, and it is called the mempool. Transactions sit in it ordered by fee rate, and they get evicted when it fills. If you want to understand that queue — its eviction rules, its 300 MB default ceiling, its 336-hour expiry — I wrote the whole thing down in what the Bitcoin mempool actually is. What you have to hold in your head simultaneously is that the queue and the clock are two independent random systems, and neither one is aware of the other.
The Constant That Makes It All Work
The “ten minutes” is not folklore. It is a number in the client:
// Bitcoin Core, chainparams.cpp
consensus.nPowTargetSpacing = 10 * 60; // 600 seconds
consensus.nPowTargetTimespan = 14 24 60 * 60; // 1,209,600 seconds
// retarget every 2,016 blocks, clamped to 4x in either direction
Every 2,016 blocks — roughly two weeks — the network compares how long those blocks actually took against the 1,209,600 seconds they were supposed to take, and scales the difficulty by that ratio. If blocks came in fast, difficulty rises. If they crawled, it falls. That is the thermostat I took apart in how the difficulty adjustment works, and its most relevant property here is that it is slow: the target you are mining against was set up to two weeks ago.
Satoshi was not guessing about the statistics either. Section 11 of the whitepaper computes the probability of a double-spend by modelling the attacker’s block arrivals as a Poisson process with expected value λ = z·q/p. The distribution was in the design from the first draft. What most readers skip is what a Poisson process implies for the thing they care about — their own wait.
P(wait > t) = e^(−t / 600) · P(next block within 10 min) = 1 − e⁻¹ = 63.2% · median wait = 600·ln2 ≈ 416 s
The Number That Contradicts the Label
Plug in the target and you get a result that surprises almost everyone the first time: if block arrivals follow an exponential distribution with a 600-second mean, the median wait is 600 × ln 2, which is about 416 seconds — six minutes and fifty-six seconds.
6.93 min / 36.8% / 4.98% —
median wait for the next block · chance the wait exceeds the 10-minute average · chance the wait exceeds 30 minutes
The mean and the median disagree because the distribution is lopsided. Most waits are shorter than ten minutes, and a small number of very long waits drag the average up. There is no such thing as a “typical ten minute block”. There is a typical seven-minute block, and an average inflated by the occasional forty-minute one.
Here is the table I actually use when somebody asks “how long does a Bitcoin transaction take”. Each probability is e^(−t/600), and the last column converts it to how many of the day’s ~144 blocks you would expect to see at that spacing.
| Wait between blocks | Probability | Roughly, per day |
|---|---|---|
| under 1 minute | 9.5% | about 14 blocks |
| over 10 minutes | 36.8% | about 53 blocks |
| over 20 minutes | 13.5% | about 19 blocks |
| over 30 minutes | 5.0% | about 7 blocks |
| over 60 minutes | 0.25% | one every three days |
Read the first row again. On a perfectly healthy network, about fourteen times a day two blocks land within a minute of each other. Nothing is wrong. That is just what an exponential distribution looks like when you sample it 144 times.

Confirmations Are a Sum of Coin Flips
One confirmation is not “the block containing my transaction”. It is “the block, plus every block after it”, and each of those is an independent exponential draw. So the wait for n confirmations is not exponential — it is a sum of exponentials, which is the Erlang/gamma distribution, and it behaves much better than a single interval.
| Target | Mean wait | Median wait | Chance you are done within an hour |
|---|---|---|---|
| 1 confirmation | 10.0 min | 6.9 min | 99.8% |
| 3 confirmations | 30.0 min | 26.7 min | 93.8% |
| 6 confirmations | 60.0 min | 56.7 min | 55.4% |
That last row is the one to sit with. Six confirmations is the industry’s default assumption for “settled”, and it corresponds to an hour on average — but only 55% of the time do six blocks actually arrive inside an hour. The other 45% of the time you are waiting longer, sometimes much longer, and none of that indicates a problem with the network, the miners, or your transaction.
Six confirmations is not a receipt. It is a probability with a widely-advertised round number attached to it. If you are timing something that genuinely needs an hour, you need a wider margin than the folklore gives you.

Three Things This Should Change In Practice
And one caveat that belongs in every article about averages: the “ten minutes” is a target whose error is corrected every two weeks. When hashrate grows fast, blocks come in a little faster than 600 seconds until the next retarget catches up; when miners shut off in a hurry, they come in slower. The distribution shape holds. The mean is a moving estimate, not a constant — which is another way of saying that even the average you are comparing yourself to is the network’s best guess from two weeks ago.
The habit worth building is small: when a transaction feels slow, ask which of the two systems you are actually waiting on. If your fee rate is above the current mempool floor, the answer is “the clock”, and the clock owes you nothing. If your fee rate is below it, the answer is “the queue”, and the queue can be bought out of — or, in the case of a stuck transaction, replaced. Those are two different problems with two different fixes, and people routinely spend money on the wrong one.
So here is the question I would leave you with, and it is not rhetorical. If ten minutes was never a schedule, what else in this market are you still reading as one?
(The End)


