How this works, and where it stops working.

What a share is worth

A bitcoin treasury company is a pile of bitcoin with claims against it. Divide the coins by the shares and you get bitcoin per share. Multiply by the bitcoin price. Subtract the debt and preferred stock that rank ahead of you. That is what a share owns.

The market almost never pays exactly that. What it pays instead, divided by that value, is the premium. It is the most important number here, because it is the part that has nothing to do with bitcoin. A company can buy more coin per share while its premium falls, and you still lose.

The premium is a multiple. Bitcoin per share is a rate. A wrapper is worth more than its coin only if it adds coin per share faster than a holder's required return. That identity converts one into the other, but cycle tops are not present values, so the calculator does not turn the Premium column into a bitcoin-per-share rate. The column is the multiple: 1.55× in 2029 on the default path. Bitcoin per share is the Premium fold's box (2% on Repeat), following the scenario unless Custom rate is on. Ticking that switch at the figure already shown changes nothing. A typed growth rate moves share prices and totals; the column stays. What the rate would be worth as a multiple of NAV is the check, not a rewrite of what cycle tops have paid. The last filing window measured -18.0% over 47 days with coins frozen: that is dilution, not a bitcoin-price effect, and it is not the dial's opening rate. The page states this beside the path. Whether a year's raise adds or costs coin per share is decided the same way: not by a chosen threshold but by the balance sheet, a raise accretes only while the premium on the path's basis exceeds one plus senior claims over the gross value of the coin, 1.23× today and falling as the stack runs down.

Convertible debt makes this circular: a note converts if the share price is high enough, which adds shares and cancels the debt, which changes the share price that decided it. The model solves that loop rather than picking a side of it.

Why selling has to clear a bar

Selling near a top and buying back lower is not free. You never sell the exact top or buy the exact bottom. On past cycles a laddered sale got about 82% of the top and a laddered rebuy paid about 1.36× the low, and there is slippage on top. Those are means: the sale kept 75%, 84%, 87% of the top at the last three tops, and the rebuy paid 1.49× and 1.22× the low at the last two bottoms.

Multiply that through and the fall has to be at least about 43% before the round trip leaves you with more shares than just holding. On the best fills on record the bar is 32%; on the worst, 52%. Shallower than that and you have paid to end up with less. The headline on the page carries the same width: the figure at the mean, and under it what the worst and best fills would have delivered.

So the model refuses to sell at a top unless the fall after it clears that bar. The bar moves when you change the fill assumptions, because it is calculated from them rather than typed in. Those fill numbers come from three tops and two bottoms, which is not many, and they are the thinnest thing on this page.

The four-year clock

Tops have landed 12, 17, 18 and 18 months after each halving. Lows have landed 26 to 30 months after. That is the rhythm the model's cycle years come from.

It is four observations. It could stop: issuance is already under 1% a year and nearly gone after 2028, so whatever the halving is doing it is running out of room to do it. The model does not assume the rhythm survives; it just uses it to place the years, and you can move them.

The scenarios are a ladder, not a forecast

There is no neutral bitcoin path. Every default is somebody's guess, so this one is written down and every figure on it can be typed over.

The rungs stretch and squash that path. They never change which years are tops, so the same years always trade and a comparison between rungs is a comparison of one thing.

These are not equally likely futures. Each cycle has topped at a smaller multiple of the last: 39×, 17×, 3.4×, 1.9×. Nothing on the record says where that stops. No weights are set here, because the only person whose weights belong in your model is you. Repeat on a holding floor is not a repeat: the floor binds, so the next top has to rise. Clicking Holds while that rung is selected lands on Smaller instead.

Each rung is a whole world

A rung is not just a bitcoin price. It also says whether raising money buys or costs coin per share, and whether the debt and preferred stack runs down. Those used to be identical across most of the rungs, which meant the rail was asking a bitcoin question on a page about a company.

They hang together on purpose, because they are not independent in life either: buying coin by issuing shares needs a premium to issue into, preferred is what you issue when common is not accretive, and premiums expand in bull markets.

One rung is calculated, not chosen

Smaller is derived. Bitcoin's bottom-to-top multiple has shrunk every cycle, so Smaller is that trend continuing one more step. Its later tops are the same shrinkage applied again, which is honest but stops meaning much: apply it a third time and a cycle top comes out below the low before it. Shallow sits below that, authored so the model's own rotation gate fails once the floor is holding.

Buying things out of a position

Most of this page is about a position. This part is about what the position is for. Name a thing, say what it costs, what happens to its value, and the years you buy and sell it. The years are yours.

There is no advice about when, deliberately. The whole effect is opportunity cost, so the answer to "when is cheapest" is always "later", and a page that said so on every purchase would just be telling you never to buy anything. Instead it runs the same plan without your purchases and shows the difference. On the default path an $850,000 house costs about $3.7M by the end if you buy in 2030, and about $1.3M if you buy in 2034. That gap is the argument. A recommendation is not.

How it is paid for, in order: borrow what the position will carry, then spend cash, then sell shares for the rest. It only refuses when all three together fall short.

The loan is sized against what the position looks like after the purchase, not before. Borrow against your shares, then sell a third of them to cover the rest, and the loan is now against a smaller pile, which breaks the margin limit and forces another sale to fix. So you get a smaller loan and no margin call rather than a bigger one that does not survive the year.

Debt is not forgiven. Selling the thing clears what it can, and a car is worth less than the loan against it, so the rest stays owed and keeps accruing. A thing you keep is still worth something, so it counts on both sides: its value and its loan.

How the model spots a bottom

Two of the old bottom signals measured the same thing: is it cheap. One compared price to a fitted line, the other to its own four-year average. Across the last four cycle lows the second was lit at all four and the first at two, and the first never caught a low the second missed. So there is one now, and it is the one that works.

The fitted line still sets the zones and the chart. It just stopped voting twice for the same fact.

The engine's own turn test, run on the record. Inside the model a year trades as a top when the premium reaches its sell level and bitcoin has stretched far enough off the lower of the last two rows; a bottom when it has fallen to 78% of the higher of them. Fed the real annual highs and lows from 2013 to 2026, in the same row convention the path uses, the bitcoin leg on its own catches every top it can see (2017, 2021, 2025) and every bottom (2015, 2018, 2022), and it also flags 6 other years that were not tops (2016, 2018, 2019, 2020, 2023, 2024) and 5 that were not bottoms (2014, 2019, 2023, 2025, 2026). On annual rows the bitcoin leg is a filter, not a detector. What decides which years trade in this model is the premium column, as written: the tops are where it reaches 1.27. Said plainly because it is where the timing comes from, and it was not measured until now.

Where the edge comes from, and when it runs out

Rotating beats holding for one reason: you end up owning more shares. Sell into a top, buy back after the fall, and if the fall was deep enough to cover the costs you hold more of the same company than you did before. Everything else is that, compounded.

Which is also why it expires. Falls have got shallower every cycle: 93%, 85%, 84%, 77%, 53%. When they drop below the bar, the model stops selling. It does not warn you, it just holds, and the page says which year that happens.

That is worth reading twice, because it sounds like failure and is not. A world whose cycles are too shallow to trade is a world where you can simply hold.

What this cannot tell you

It cannot time anything. It places tops in the years the clock suggests, and if the real one comes a year early every figure moves. How much is on the page under Against the alternatives, as two runs beside the on-time one: every signal a year early, and every signal a year late, into the fall. On the shipped record at the default dials, a year early delivers 10% of the on-time figure and a year late 62%.

It cannot see a company doing something stupid, or a rule change, or a hack. It has no view on tax beyond a rate you type. And its execution numbers come from five events.

It is a calculator for one question: given a path, does selling tops beat holding, and what does that fund. Everything it says is downstream of a path you can change.

Two ways a bigger buyer arrives

The Bigger rung assumes a buyer who is not in the market today. There are two roads and only one needs anyone's permission.

The first is a rule change: money that legally cannot hold the asset now can. That either happens or it does not.

The second is already moving. As falls get shallower the asset becomes better collateral, so more can be borrowed against it more cheaply, and the buyer arrives holding borrowed money instead of a new mandate. That road is measurable today.

They are two routes to the same rung, not two rungs stacked, and the second is bounded: run through this model's own falls it grows credit about 1.4× over the horizon, not the 2.5× the argument is usually told with.

The par-anchored preferred

A preferred share anchored to a $100 par is a different instrument below par than above it. Below, the company faces a permanent step up in what it pays. The page watches that line.

Where the default path came from

The first row is not a forecast. 2026 is mostly behind us and the low printed at $58,525 on 30 June, so the row sits at the market rather than at a guess made before any of it happened.

The rest is halving-timed, with tops and the bear the year after each. Every row is that year's expected high, or its expected low in a year the price falls. It is one person's reading, it is in the code rather than hidden, and if you disagree with it that disagreement is the most useful thing you can do here. The page takes it: under the path table, Move a year opens the written rows to typing, a bitcoin price or a premium for any year the path was written for. The scenario pills, the horizon and the Holds world apply on top, so the ladder runs on your path rather than around it, and the moved years travel with the link. A year moved so far that its top is no longer the top is not re-timed for you; the run simply does not trade there, which is what the path you wrote says.

The floor, and the ceiling

Bitcoin's cycle lows have risen every time. Draw a line through them and you get a floor. Draw one through the peaks and you get a ceiling. The gap between them is the cycle.

The floor here is built from the steps between lows, not from a curve. The last two steps were 4.95× and 3.71×. Each step is smaller than the last, so the next one is smaller again, and the floor rises by less each cycle.

Why not fit a curve? Because a curve needs a start date, and nobody can say what it should be. Start at bitcoin's launch and the 2045 floor is $3.4M. Start four years later and it is $682k. Both fit the data equally well. A number that moves by five times on a choice nobody can justify is not a floor. The curve is still fitted, as a check. The two still agree at the next low; by 2037 the fitted line sits a quarter above the live floor. The chart draws the live floor: the last two steps.

Four points, or five thousand? The floor is fitted to 4 cycle lows, which is a claim about four dated events. The same shape fits 4,987 daily closes instead, placed so that 5% of them close below it. That line carries an exponent of 5.71 against this floor’s 5.41. Different estimators, not meant to match, so the agreement is worth no more than corroboration. It is here because a four-point fit landing somewhere else entirely would have been a warning, and there was previously no way for that warning to reach the page. The daily record runs to 2026-08-27, as of 53 days ago; the daily record is more than 30 days old; it is imported by hand, so a gap here is a gap in the import and not in the market.

Is the floor-held world the optimistic one? Not on the coin price, and it is worth saying rather than leaving to assumption. The bound the ladder rests on when the power law holds is $64,965 at this year end. Only 1% of days since 2013 have closed below the line now at $72,070, and the bound sits under even that. So the floor that world holds to is cheaper than a level the record has been beneath on about one day in 100. Where that world is bullish is the slope it implies decades out, which is what the share-of-world check is for. Nothing is clamped to any of these lines.

Does the line predict, or only describe? Every other figure here is measured on the same lows the line was fitted to, so it is bound to look reasonable. The test that is not: fit the line on the first 3 lows only, then ask it for the 2026 low it has never seen. It says $60,367. The low was $58,525, 3.1% below it. That is one prediction off three points, which is thin, and it is not a claim that the next one lands as close. It is here because it is the only check on this page that the floor can fail while every other number about it stays healthy. The method is @sminston_with's, who freezes his own power-law fits at an early date and scores later prices against them.

The floor uses every low since 2013. The ceiling uses the last three peaks. A bottom is where capital that sees undervaluation has shown up; that record can be longer. A top is momentum, and old mania is a poor guide to the next one. The 2011 cycle is still out of both: that was a different market, not a different kind of signal. The next-step on the floor still uses only the last two rises (4.95× then 3.71×), because the 18× jump from 2015 to 2018 has a mania in the middle of it.

Why a rolling window at all, then? The market keeps changing, so old peaks stop being evidence of how wide the next one will be. The cost of a short top sample is named: three peaks is not much data. The cost of throwing away a mature low to stay in lockstep with those peaks was the worse one, and it is the one this no longer pays.

The ceiling is the floor times a shrinking multiple. Cycle peaks have run 18×, 12×, 6× and 2.7× above a line through the lows. The channel is closing. It closes without ever meeting the floor, because what shrinks is the gap and a shrinking gap never reaches zero. A straight ceiling would cross under the floor instead, which is why it is a curve.

Pick "Power law: Holds" and both edges bind. Lows sit on the floor, peaks stay under the ceiling, and the cycle is whatever fits between them. Pick "Fades" and neither does: the lows fall through and the path is a forecast instead. The page opens on Fades, because opening on Holds would open on a floor that already asks for a large share of the world's money by the end of the written path.

Where the live pair comes from

The calculator paints with the pair last published here. It then asks CoinGecko for bitcoin and the share, in one request, so the premium is two numbers from the same moment. If that request fails, the published pair stays. CoinGecko sees that you loaded the page; nothing you type is sent with it.

Where this stops working

A shallow cycle is not worth trading. Selling a top and buying the low back costs something, so the fall has to be big enough to pay for it. Around 43%. When falls get shallower than that the model stops selling, and that reads like the strategy failing. It is the same fact as: the cycles got small enough that you can just hold.

The floor cannot rise forever. If bitcoin ever became the money everything is priced in, there would be nothing left to price it against, and the line would stop meaning anything. So the page checks the floor against the size of the world. It stays inside it at every horizon. On the old version, which started the clock at launch, it did not. The path in view is checked the same way, because the floor is the same on every rung and the path is not: each row's price times 21M coin, against broad money of about $120T today growing 7% a year — about the compound rate of global broad money since 2000, not an accelerating rate and not a softer round five. Past about a quarter of that (24% on the page, roughly the share gold holds), the lead tile says what the figure is asking for and the rows from that year are dimmed. Nothing is capped; a long horizon on a high rung still prints its number, beside a sentence about what the number means. The stock is stated, not sourced; the growth rate is the recent record. The conclusion is not sensitive to a factor of two in the stock.

None of this is a law. It is four cycle lows since 2013, three peaks for the channel, and two steps for the next rise. It says lows have risen and peaks have got relatively smaller. It does not say by how much they will next time, and anyone who tells you it does is reading a fitted line as a prophecy.