Why Blocks Take 2 Minutes or 40

Dice, not trains

At network scale, nearly a sextillion fingerprints are checked every second, each one an independent try with the same vanishingly small chance of landing under the bar. Nobody is scheduled to win; a block simply happens, somewhere on Earth, whenever one try clears. Finds land like raindrops on a roof — steady on average, patternless in detail — not like trains on a timetable.

That one sentence decides everything else on this page. A timetable has memory: a late train means a crowded platform and a conductor making up time. Dice have neither. The 10-minute pace is enforced only on average, by the difficulty thermostat at its fixed checkpoints — never gap by gap. Statisticians have names for what falls out of independent tries at a fixed bar — the arrivals form a Poisson process, the waits between them are exponential — and having named it once, this lesson will go back to calling it dice.

Most gaps are shorter than the average

Now the part that surprises people who have stared at the advertised number for years. Our node has banked every consecutive-height BTC gap since December 2019 — 322,802 of them. The average gap: 594.5 seconds, or 9.9 minutes — on schedule on average, historically a touch fast because hashrate tends to grow mid-epoch and the thermostat corrects after the fact (the mechanics are the difficulty lesson’s turf). But the typical gap — the median, the one in the middle if you line all 322,802 up — is 413 seconds. Under seven minutes. 63.5% of all banked gaps beat the 10-minute figure; half of all waits are over in 6 minutes 53 seconds or less.

The average is not lying — it is being propped up. Short gaps are the most common thing the network produces, and the pile of them is balanced by a thin tail of brutal droughts that drags the mean back up to 10 even though two-thirds of blocks arrive early. The shape even comes with a built-in receipt: for dice-shaped waits the median must equal 0.693 × the mean. Ours: 413 ÷ 594.5 = 0.695. At a true 600-second average the predicted median is 416 seconds; we banked 413. Theory and seven years of receipts land three seconds apart.

The tail, counted: pure dice vs our archive

How often do the brutal gaps land? Pure dice make an exact prediction with zero knobs to tune: the share of gaps running past k× the target is e^−k — that is the entire formula, nothing fitted to the data. Here it is next to what our node actually banked over seven years:

Banked column self-computed 2026-10-02 from 322,802 consecutive-height BTC gaps (Dec 2019–Oct 2026), orphans and archive-hole boundaries excluded. The prediction column is e^−k arithmetic and nothing else.

The dice called it to a few tenths of a percent, across seven years, sight unseen. Read it as odds: about one gap in twenty runs past half an hour, one in fifty-six past forty minutes, roughly one in 375 past a full hour. That last one sounds rare until you do the blocks-per-week arithmetic: at 144 blocks a day, an hour-long drought lands somewhere in the world about two or three times a week, forever — our archive holds about 860 of them. The longest it has ever banked: 8,354 seconds — 2 hours 19 minutes — before height 689,301, on 1 July 2021. Hold that one; it comes back below.

And the shape does not age: the most recent 13,630 gaps alone — the archive’s contiguous live edge — bank 5.2% past 30 minutes and 1.9% past 40, median 423 seconds. Same dice the full seven years show.

The dice have no memory — network edition

Module 1 settled the personal version at the machine: six dry hours change nothing about your rig’s next pull — that lesson owns that question. The network edition is the one that fools dashboards. A 40-minute drought just ended. Is the next block “due”? No. The expected wait, measured from this exact second, is the same ~10 minutes it was an hour ago, a week ago, and in 2019. Three long gaps in a row predict nothing about the fourth. When a feed looks slow for an afternoon, you are almost always watching a few bad rolls land in a row — and nothing else.

One stranger consequence, worth exactly one paragraph: open the live feed at a random moment — you land somewhere mid-gap — and your expected wait to the next block is not five minutes but the full ten. The dice do not know how long the current gap has already run (that is what memoryless means), and random moments land disproportionately inside the long gaps, which are exactly the ones with room to land in.

Scatter is luck; a sustained change is hashrate

So split every change of pace into the only two kinds there are. The scatter — everything above — is luck, is permanent, and is identical at any hashrate: more machines make the clock tick faster, not steadier. A sustained change of pace is different: machines joined or left, the bar is now wrong for the fleet pulling at it, and the retarget absorbs it at the next checkpoint — concept only here; the machinery is the difficulty lesson’s. The thermostat fixes the average and never touches the scatter, because the scatter was never broken.

Our archive holds the cleanest possible receipt on the difference. That record 2h 19m drought sits in mid-2021, in a stretch when a large share of the network’s hashrate went offline mid-epoch. Fewer machines pulled at a bar set for more machines, so every gap stretched — the whole distribution slid right, which is exactly what sustained change looks like — until the checkpoint landed two days later, on 3 July 2021, and difficulty took its largest step down on record: −27.9% (recomputed from our own headers table: difficulty 19.93 trillion → 14.36 trillion at boundary height 689,472). The pace snapped back; the scatter stayed, as it always does. Watch today’s bar move at /btc/difficulty.

Same dice, tighter clock: Litecoin and Dogecoin

Nothing above is Bitcoin-specific — the dice only care that tries are independent and the bar holds still between adjustments. Litecoin plays the identical game at a 150-second target, Dogecoin at 60. Tighter clock, same shape:

Self-computed 2026-10-02 from our own litecoind and dogecoind header archives, same method as the BTC table: consecutive heights, orphans excluded.

Litecoin is the clean second confirmation: 663,631 banked gaps, median 103 seconds against a predicted 104 (0.693 × 150), and 4.96% of gaps past 3× target against the predicted 4.98. Two chains, two archives, one shape.

Dogecoin prints the honest exception: median 43 seconds (prediction 41.6), but a measurably heavier tail than pure dice — 6.08% of gaps past 3× target against 4.98 predicted, and 2.82% past 4× against 1.83. We print that as a measurement, not a story. Our best reading of the cause — hedged, not asserted — is its per-block DigiShield retarget oscillating under merged-mining arrival patterns; and a 1-minute clock magnifies miner-clock noise proportionally, so some of the extra tail may be timestamps wobbling rather than dice misbehaving. Watch the same Interval column run ten times faster at /doge/blocks, or at Litecoin pace on /ltc/blocks.

What a “gap” even is

Honesty footnote before you watch anything live. Every “gap” on this page is the difference between two miner-reported header timestamps on consecutive heights — and miners’ clocks wobble. 0.57% of our banked BTC gaps are negative: a block stamped earlier than the block before it (Litecoin and Dogecoin: 0.002% and 0.003%). Nothing is wrong — the protocol tolerates loose clocks — but it means an Interval column is a good ruler and a bad stopwatch, and per-gap numbers carry a little timestamp noise on top of the dice. Our stats keep it honest the boring way: consecutive heights only, orphans excluded, archive holes excluded, every figure dated and carried with its gap count. The previous lesson handed this page its parting line — the wobble in hashrate charts is mostly luck wearing a chart costume — and this page is that luck, measured.

Watch the dice wobble, live

The interval tile our dashboard wires into these lessons is itself a small exhibit of everything above: our nodes average the last 200 header gaps per chain and compare the result to target.

FAQ

Why do some Bitcoin blocks take 40 minutes or even an hour?

Because block discovery is dice, not a schedule. Every hash is a fresh independent try, so the gaps between finds follow a skewed distribution: from 322,802 gaps banked by our own node since Dec 2019, about 4.8% ran past 30 minutes, 1.8% past 40, and 0.27% past an hour — almost exactly what pure dice predict. Nothing is broken, nothing is slow; an hour-long gap lands somewhere in the world roughly 2-3 times a week, forever.

Is Bitcoin’s block time exactly 10 minutes?

No — 10 minutes is a long-run average, not a metronome. Across our archive the mean gap is 9.9 minutes (historically a touch fast, because hashrate tends to grow between retargets), while the typical (median) gap is just 6.9 minutes. Individual gaps swing from seconds to over two hours; the longest our node has banked is 2h 19m, on 1 Jul 2021.

Why is the median block time shorter than the average?

The distribution is lopsided: lots of short gaps, a thin tail of brutal ones. In our archive 63.5% of all BTC gaps beat the 10-minute figure, and half of all waits are over in under 7 minutes — but the rare 30-and-60-minute droughts drag the average back up to 10. The math behind it (median = 0.693 × mean for this kind of arrival process) matches our banked data to within three seconds: 413s observed vs 416s predicted.

After a long gap, is the next block more likely to come quickly?

No. The dice have no memory at network scale: a 40-minute drought tells you nothing about the next gap, whose expected wait is the same ~10 minutes it always was. If blocks stay slow for thousands of blocks in a row, that is not luck any more — that is hashrate leaving, and the difficulty retarget absorbs it at the next checkpoint. Scatter is luck; sustained change is the thermostat’s job.

How do Litecoin and Dogecoin block times compare?

Same dice, tighter clock. Litecoin targets 2.5 minutes and our 663,631 banked LTC gaps match the dice prediction as cleanly as Bitcoin’s (median 103s, 5.0% beyond 7.5 minutes). Dogecoin targets 1 minute (median 43s) but runs measurably streakier than pure dice in our archive — 6.1% of gaps beyond 3× the target instead of the predicted 5.0% — a real quirk of its per-block retarget and merged-mining arrival patterns, printed here as a measurement, not a story.