Every Metaplanet share owned about 19 percent less Bitcoin than my number said, and my number was built to catch exactly this kind of thing.
That sentence took me a week to write. This piece is that week. I spent it looking for what was hiding, working out how it hid from a framework designed to find it, running a fleet-wide census, updating the CEBE standard for this instrument class, and building a tool that now sits on the site as a result. I am not writing it up because the story flatters the framework. Parts of it do not, and the piece doesn't exist without those parts. Growth is not only watching a sector adopt your approach. It is what you do when the error is in your own work.
You own it as loudly as you were wrong. You detail exactly how it resolves. You show what was learned from the experience.
The instrument
Metaplanet carried a series of stock acquisition rights, the 10th Series, granted in December 2022. The terms are filed in the FY2025 annual securities report (EDINET S100XTWY, filed 2026-03-26), and one clause is the catch. The adjusted share count under the rights equaled the fully diluted share total multiplied by 0.2. The pool was contractually one fifth of the fully diluted count, it adjusted whenever that count changed, and the clause carried no cap. The strike is 10 yen and the stock closed at 220 yen on the TSE on 2026-06-26. The rights are roughly 22 times in the money, the strike is about 4.5 percent of the share price, and the exercise period runs to February 2033. These rights would have been exercised. Hold the past tense in those sentences. It became new today, while I was finishing this article, and the ending will come back to it.
I want to highlight something I think is important in understanding these counts. Fully diluted shares is the instrument's own defined term inside its filing. This is not the issuer's published effective diluted metric and not an accounting figure. I use each count by its own name but they are not the same construction and not consistent between treasuries.
Why my number missed it
CEBE is built from the common shareholder's seat. Take the Bitcoin, subtract every senior claim at its accrued value, divide by basic shares. Debt is a claim, preferred equity is a claim and convertible debt is a claim until the moment it becomes share. The framework argues that senior claims are what the popular per-share metrics ignore, and that basic shares are a fact while every fully diluted count is a model with the issuer controlling inclusion.
The 10th Series walks straight through that construction, because it is not a claim. Nobody is owed a yen. Nothing accrues, nothing matures and nothing stands ahead of common in a liquidation. Their entire effect arrives on the other side of the division sign. When they exercise, the denominator grows by a fifth.
So the framework reported the treasury correctly, the claims correctly, and the shares that exist correctly. A reader still walked away knowing less than the situation warranted, because a deterministic 20 percent share expansion sat outside all three terms. My number was right for what it measures. The number was also not enough.
The concession
Here is the part I want to say plainly, unburied. The diluted counts I have spent months criticizing flagged this instrument the whole time, and my basic count did not. At Metaplanet that count is the issuer's own defined term, effective diluted shares outstanding, built by the issuer's own stated method.
That deserves an honest accounting without being defensive. The diluted count caught the pool because the issuer's inclusion rules admitted it. Once an instrument is admitted, the blend carries it whether it is 22 times in the money or hopeless. The composite hands you a single number with no attribution, so a reader staring at Metaplanet's diluted count still could not tell you the 10th Series existed, what its strike was, or that this one instrument was about 20 points of the gap between the counts. The number was right and uninformative at the same time. Hold the word admitted. Not for its tense, but for what it says about who controls the door on every dilution model.
My miss was a different kind of failure. The basic-shares architecture was not wrong. My instrument coverage was incomplete. There was no disclosure surface for things that deliver shares without ever being claims, so this one sat outside every row I created to track these treasuries. The fix was not to adopt the blend. The fix was to build the row I was missing.
The search
Once one company shows a proportion-defined pool that my coverage failed to capture, the only responsible next question is who else has one. So every company on the tracker went through the same census, every instrument that adds shares to the denominator without ever standing in the claims stack: warrants, rights, options and RSU programs, share-settled obligations, contingent issuance, read from the filings. My earlier sweeps had hunted predatory mechanics. This sweep hunted denominator mass because the Metaplanet pool was not predatory. It was just large, cheap, and invisible to a claims lens.
The census came back from thirteen companies. Proportion-defined pools exist at six of them.
Metaplanet's 10th Series is the specimen. The Smarter Web Company's incentive plan defines awards as a percentage of issued capital across twenty milestone tranches, the same proportional sizing at smaller scale but behind far harder gates. Capital B holds a 300 million euro issuance authorization with no share count at all, which at the 2026-08-03 close corresponds to something near twice the current basic count. American Bitcoin's equity plan replenishes itself to 20 percent of fully diluted every year through 2035. Nakamoto's plan does the same at 5 percent annually. DDC's warrant program defines board-issuable warrants at 25 percent of outstanding shares with no stated ceiling, alongside an employee plan that tops itself up annually under a 15 percent aggregate cap.
It is worth sitting with what proportion-defined means. In plain terms, when they issue more shares, they are allowed to issue more shares.
Three more companies carry large fixed pools rather than proportional ones, and one company appears in both lists. ProCap's market-based RSUs and its seller earnout share a single 9 dollar price trigger against a 1.71 dollar stock, so they sit dormant. Strive carries a warrant wall around 31 percent of basic, far out of the money. OranjeBTC carries the largest latent block in the census, a second class of stock convertible at a fixed ten for one ratio into roughly ten times the traded float, with the bulk of it callable by the company itself for about 3.7 million reais. And Smarter Web, already counted above for its proportional plan, separately carries fixed count warrants at 2.5 pence, around fourteen times in the money and about 9 percent of the share count. The remaining four companies came back clean or minimal, and those attestations are worth as much as the findings, because now the absence is verified in the census record rather than assumed.
What the diluted count cannot see
The most important finding of the census was not any single pool. It was a measurement about the fully diluted counts themselves, and three companies demonstrated it three different ways.
Metaplanet's published effective diluted count reconciles to the share. By the issuer's own footnote, that count excludes the moving-strike warrant series, 428,042,000 shares, about 33 percent of the share count, lawfully and by stated method. This is also a different construction from the fully diluted term defined inside the instrument's own filed conditions, two objects at one issuer under one phrase. And the pattern is not Japanese. Strategy files two diluted labels with two different definitions, and the one its offering decks carry as the headline counts only the instruments at or below the market price. Capital B's diluted figure excludes roughly 40 percent of basic plus the unbounded authorization, again disclosed. American Bitcoin's largest dilution mechanism, the ungranted plan reserve, is invisible as constructed, because accounting standards only count instruments that exist, and a reserve is permission for instruments to exist later.
None of this is concealment. Every exclusion is disclosed and every method is defensible under its rules. That is exactly the point, and it is where the concession earlier in this piece gets its correction. I credited the diluted count with catching my miss, and the credit stands, but the catch was not coverage. It was admission. The same construction that admitted the 10th Series excludes a larger pool at the same company by stated method. The count is issuer defined, and the definition is the inclusion rule. So the sentence I would put on a wall:
A clean diluted reconciliation is evidence only that the instruments the issuer chose to include are accounted for.
I used to read the diluted count as the conservative number, the company showing me the worst case. It meant all companies accounted fully for the dilution potential and that a fully diluted count was comparable company to company. Working through this fleet, it can no longer be read that way. The diluted count is a summary governed by inclusion rules, and the rules differ by issuer, by jurisdiction, and by instrument class. Sometimes the worst case is in the summary. Sometimes a third of the share count is standing politely outside it, in a footnote, excluded by method.
The tool
The site now carries a disclosure built for this class, and it is deliberately not a new blended number. It is a bracket.
For a share-delivery instrument, there are exactly two states supported by the filings. The standing state which are the shares that exist today, basic. The as-exercised state: the same snapshot recomputed with the pool delivered, the strike proceeds added to cash, and the claims untouched. Both states are computed live from filed terms. The width between them is the envelope, and the envelope is the disclosure.
This is not a novel invention, and that is its strength. It is the pro forma discipline that audited financials have carried for decades. It is the same split accounting standards already make between basic and diluted earnings per share, two lines, deliberately never merged. The presentation is borrowed from conventions every filing reader already trusts.
The weight was always there
None of this is invented here, and saying so makes the adoption of the change stronger. Traditional finance has weighted share delivery by moneyness for as long as diluted counts have existed. The treasury stock method is a moneyness weighting with exactly one step. An out of the money option counts for nothing, an in the money option counts in full with its strike proceeds offsetting the count. Options desks price the same question continuously, hedging dilution at delta rather than at face count. Stock compensation accounting expenses grants net of expected forfeitures, with exercise behavior estimated and calibrated against history under audit. Earnout accounting carries contingent shares at probability weighted value, a gate treated as a probability rather than a strike distance.
To adapt, the framework is now doing to the denominator what it did to the claims side. Every share delivery instrument carries a stated delivery weight from a published prior, banded by moneyness and gated by vesting and contingency, per instrument. The priors are public, the bands are challengeable, and the calibration study that tests them against filed exercise outcomes is pre registered before its data is scored. Until that study clears, no weighted denominator figure renders anywhere. The unweighted bracket above remains the published disclosure, because its endpoints are arithmetic on filed terms and a weight is a forecast.
The record
One more census result belongs here because it cuts in common equity's favor, and verified absence is worth as much as any finding. Nakamoto's filings once authorized up to 600 million pre-split shares under a marketing services agreement, a figure that would have dwarfed every pool above. The census confirmed the agreement terminated when the counterparty became a wholly owned subsidiary in February 2026, and the residual authority is extinguished.
So that was my week. After my framework missed an instrument class the standard matured and my analysis sharpened. The distinction between those two words is the real lesson. The framework is how I see: the common seat, the claims netted, the shares that exist. The standard is what I have written down so far, versioned and amendable, and the difference is why one can hold while the other grows a row. Frameworks are judged by whether they were wrong. Standards are judged by whether they are complete. This week the framework held, and the standard was incomplete, and now it is less so. The gap exposed is now a disclosure surface with filed citations behind it. The search that followed found the class at nearly half the fleet, reinforcing my argument for an issuer agnostic standard. It also corrected my own results twice along the way, once in my favor and once against. Every figure above carries its date and its source.
On 18 August 2026, as this piece was being finished, Metaplanet's board eliminated the Adjustment Provision by resolution, effective immediately, and fixed the pool at 319,464,000 shares with a five year lock up (TDnet 140120260818522530). The clause that adjusted the count whenever the count changed is gone. The envelope this piece measures, published on the tracker in the days prior, now describes an instrument that no longer moves. This piece claims nothing about why. It notes the dates, because that is what a measurement does.
The pool taught this lesson at the denominator, but it is not a denominator lesson. It is the same rule at every layer of this work, and it is published in the methodology. I would add that life brings its own influence here. When faced with the choice between a quiet correction and a loud failure, I will take the loud failure every time. Quiet corrections teach you how to disguise errors from view. Loud ones teach you where errors hide.
A number that will not tell you what it contains has already told you something.
The headline stays basic. The bracket shows the rest.