Bitcoin: the cycle clock

What the halving really does, why the models that explain it fail,
and where the bottom lies when you stop looking at price and start looking at time.
3 August 2026 · all measured, all repeatable

1. What the halving really does — and what it does not

Every 210,000 blocks, the reward miners receive for each block is cut in half: 50 → 25 → 12.5 → 6.25 → 3.125 bitcoins since April 2024. That is all the halving does. Not one thing more.

The clock is not the calendar. It is the block.

It is repeated everywhere that the halving happens «every four years». That is not true, and understanding why is the basis for everything else:

What Bitcoin regulates is not the date: it is the difficulty. Every 2,016 blocks —about two weeks— the network looks at how long those blocks really took and adjusts the difficulty so that the next one takes ten minutes on average. The «four years» are an arithmetic consequence of that: 210,000 blocks times 10 minutes is 3.995 years.

But the adjustment always arrives two weeks late. If computing power grows fast, blocks come out in less than ten minutes for that whole fortnight and the halving arrives early. That is why the 2016 one came four and a half months ahead of the theoretical count: in those years computing power was growing at rates of +275% and +475% a year.

And now the number that takes apart the usual explanation

The story goes: the halving cuts new supply, scarcity pushes the price up. Let's put figures on it.

2. The awkward question: if it gets scarcer, why does it fall first?

If scarcity pushed the price up, the logical thing would be for it to rise the day after the halving. It does not. It takes between twelve and eighteen months to reach the top, and along the way it usually falls. This is what looks illogical — and it is not. What is illogical is the explanation it is given.

First: the halving is known years in advance

Down to the exact block. Anyone can count it. If the market were even minimally efficient, it is already in the price before it happens. There should be no effect the day after, and there is none.

Second: right after the halving, the opposite of what was promised happens

The miner sees their income cut in half from one day to the next, while the electricity bill stays the same. The least efficient sell reserves to pay it, and some go bust. In 2022 Core Scientific fell —it was worth more than 4 billion dollars—, as did Compute North, and almost Argo Blockchain.

Third: the rhythm of the cycle does not fit a supply shock

A supply shock would be instantaneous. What we observe —twelve to eighteen months of rise, euphoria, and then a year of decline— fits something completely different: it rises, draws attention, new money comes in, it rises further, leverage appears, the marginal buyer runs out, forced liquidations arrive and it falls in cascade.

3. Why all the famous models have broken

Stock to flow: the scarcer, the more expensive. It worked for years. From 2021 it drifted: it forecast half a million dollars per coin and the actual high was $124,753.

Rainbow chart: it splits price history into coloured bands over a logarithmic curve. It worked, until the price fell out below it and more colours had to be added to make it fit again. A model whose suit has to be let out has stopped being a model.

Power law: the most solid. Price inside a mathematical channel while the network grows. That channel held for fifteen years and broke in July 2026.

4. The idea that replaces them: not a number, a crossing

If no single variable explains the cycle, the way out is not to look for a better one: it is to calculate the same thing by different routes and look at where they cross. When several routes that do not talk to each other point to the same zone, that zone deserves attention. And when they contradict each other, the contradiction also informs: it tells you which one is breaking.

We call that an action area: not a point, but a band of dates and a band of prices, with the uncertainty on the table instead of hidden.

5. Route 1 — Time

The price of each cycle does not look like the previous one. The timing does.

What it says for this cycle: top on 6 October 2025, plus the 52 to 58 weeks of the complete cycles → bottom between 5 October and 20 November 2026.

6. Route 2 — Depth

Its variability (5.3%) is similar to that of time (6.1%). They tie on consistency. And they do not tie on anything that is useful:

And on top of that the number is shifting: −84.7% → −83.4% → −76.6%. Each winter shallower than the one before. A parameter that drifts is no good as a trigger. As confirmation, yes.

7. Route 3 — The previous top as a floor

In the last two cycles the bottom ended up just below the previous top. In 2018 that did not happen at all. That is two data points, and one flatly contradicts them. It is the most fragile route.

If it repeated: bottom between $54,729 and $58,581.

8. Route 4 — The path already travelled

Comparing the four winters on the same calendar day —301 days after the top, which is where we are—:

Two readings, and both matter.

The big drop happens early, not at the end. At this point the three previous winters had already done between 78% and 94% of it. What remains after day 301 is not a collapse: it is a crawl. The idea that «we haven't seen the worst yet» is not supported by history.

And an estimate of the bottom comes out of that. If the current −53.1% is 78-94% of the total path, the final drop would be −56.5% to −68.1%:

9. Route 5 — On-chain signals, and what comes out when you measure them

There is a popular model that looks neither at the calendar nor at price, but at people's behaviour. It requires four conditions at once: capitulation (less than 50% of supply in profit), an upward crossover of the hash ribbon, price below the average cost of recent buyers, and long-term holders accumulating again. It would have happened five times in fifteen years, with an average +130% at twelve months.

And the first signal, measured, has not happened

MVRV —market value divided by the value at which the coins last moved— below 1 means the average holder is at a loss. That is capitulation.

The three previous winters bottomed with MVRV well below 1. This cycle has not gone below 1.103 on a single day.

The same calculation gives the overall realised price —the average cost of all holders—: $52,788. It is the price at which MVRV would be exactly 1. Today the market is +20.2% above it, and no cycle has bottomed above that line.

10. What artificial intelligence is changing right now

This was not in any previous cycle and it is probably the most important structural change since exchange-traded funds came along.

Three quarters in a row falling. −30.8% from the October 2025 high. It had never happened before.

What explains it is that capacity is moving to another business. More than 70 billion dollars in contracts announced in the sector: IREN with Microsoft for 9.7 billion, Hut 8 a 9.8 billion lease over fifteen years, Core Scientific with CoreWeave for more than 10 billion.

The irony sums it up: Core Scientific went bust mining bitcoin in 2022 and today is worth more hosting artificial intelligence.

Five consequences, and none is what it seems

• The miner stops being a forced seller. Whoever bills Microsoft pays for electricity with the contract and can keep their bitcoin. A permanent source of selling pressure is removed.
• Miner capitulation stops working as a signal. If the marginal miner is pushed out not by price but by the profitability of another business, the hash ribbon stops measuring pain and starts measuring migration. And that is more dangerous than breaking: it keeps giving signals, but about something else.
• When a miner switches off because of price, they come back. When they sign for fifteen years, they don't.
• Mining moves to the cheapest energy —flared gas, surplus hydro— and that, contrary to appearances, makes it more resilient: it becomes the buyer of last resort for the energy nobody wants.
• Less computing power means less security and more concentration. It does not move the price tomorrow; it is the real ten-year risk.

And does this delay the 2028 halving?

About 89,600 blocks remain. Simulating hash rate scenarios:

Days, not months. The delay does not come from the hash rate being low, but from it falling: if it stabilises at any level, the difficulty adjusts within two weeks and the block goes back to ten minutes. The time anchor is more robust than one might fear.

11. Where they cross: the area

12. The economics: what is earned, and what no longer is

Here is the number almost nobody puts on the table.

And the same from the bottom to the next top, which is what is really earned: ×109.5 → ×20.9 → ×7.9.

The scenarios, with price and date

And what would be earned in each one

From that bottom to the top of the next cycle, which the geometry places around 2029-2030. The top is estimated by extrapolating the multiple, which has three readings: it is the weakest part of the whole document, and we say so.

13. How to act on an area and not on a point

If you had an exact point, you would buy there. Since what you have is an area, the way to act changes:

• You spread it, you don't fire it all at once. The money is split into tranches across the band. If the bottom comes earlier, you are already partly in; if it comes later, you still have ammunition.
• The brake cannot hang on a percentage drop. We have measured it: a tranche plan by depth —at −60%, −70%, −80%— would probably never be triggered in this cycle, and would leave three quarters of the money idle waiting for an abyss that no longer comes.
• A threshold with meaning is better than a round number. The realised price is not a parameter chosen because the backtest looks pretty: it is the exact line at which the average holder goes into loss, and it has been crossed at three out of three bottoms. A round −50% means nothing; MVRV at 1 does.
• And you need an escape valve. If the cycle has changed and that line is never crossed, the money cannot stay waiting forever: after a period beyond anything known —the three winters lasted 364, 366 and 406 days—, it goes back to the normal flow.
• The area narrows over time. In October we will know more than today. A plan that is not updated is not a plan, it is a bet.

14. How to check it: what would break each route, with a date

A method that cannot be broken is not a method. Failure conditions, written in advance:

And the joint test: if 1 December 2026 arrives without a low below $58,559 and with the price above 75,000, the bottom was in on 30 June and all four routes failed at once. It will be said just as clearly as it is said now.

15. Risks, and what this is not

• These are three cycles. Any regularity measured over three cases may be chance, and the previous-top route rests on two.
• The current cycle has not ended, so everything calculated with it is contaminated by not knowing how it ends. That is why there is no parameter optimisation here at all: when we tried it, the result «improved» the later you bought —up to a multiple of 593— simply because the price is low today. That was not a finding: it was the unfinished cycle sneaking in through the back door.
• The routes are not independent. They are five ways of looking at the same three cycles. Their agreeing is less impressive than it seems; if the pattern has broken, they all fall at once.
• The 2013 data is approximate, from the time when each exchange gave a different price.
• It is not a price prediction; it is an estimate of when much more than of how much. It is not a signal: we do not say buy or sell. It is not financial advice: it is training material, and none of these routes replaces your judgement or your tolerance for losing money.

16. And now that you have made it this far

Four thousand words to estimate when. And none of them contains the decision that really ruins people, which is how much.

You can call the cycle bottom right and still lose everything. All it takes is having put in more than you could afford to lose. No model fixes that mistake —not this one either— because it is not a market problem: it is a problem of bet size. And it is decided beforehand, with a cool head.

None of the three links is a recommendation to buy or sell anything. It is still training material, and the risk of losing capital —in part or in full— is still yours.

All measured on 3 August 2026 on daily BTC-USD closes and on-chain data from the free tier of CoinMetrics. Where a figure could not be confirmed, we say so. Sophronepsis — training content, not financial advice.