How to Read a 10-Q for Bitcoin Treasury Numbers

CEBE needs five inputs: bitcoin held, debt, preferred, cash, and shares. All five are in the quarterly filing. So are five convincing decoys sitting directly next to them, and in most cases the decoy is the number that appears first, in the larger typeface, on the page you land on.

This guide walks through where each input actually lives, what sits next to it, and why the difference matters. We'll use Strategy's Q1 2026 10-Q throughout.

Filing: Strategy Inc, Form 10-Q for the quarter ended March 31, 2026. Filed May 6, 2026. Accession 0001050446-26-000031.

Pull it directly from EDGAR:

The Archives pattern is /Archives/edgar/data/{CIK}/{accession-no-dashes}/. Strategy's CIK is 1050446 (cover page, "Entity Central Index Key"). The primary document in that directory is mstr-20260331.htm. Every company follows the same pattern; only the CIK and accession change.

Note numbers in this filing:

Note numbering shifts between companies and sometimes between quarters at the same company. Find the note by title, never by remembered number.

Step 1Bitcoin Held

Where: Note 3, Digital Assets. First table in the note, captioned as a summary of holdings "as of" the two balance sheet dates.

Note 3 gives you three lines and one of them is the answer:

CEBE is denominated in bitcoin, so the count is the input and the dollar figures are context. But you need to read the dollar figures anyway, because they are where the quarter's most dangerous error lives.

The as-of trap

Cost basis is not a market price. It is never a market price. It is the sum of what the company paid, accumulated across every purchase since inception, and it has no relationship to what bitcoin is worth on the balance sheet date.

Do the division on this quarter and the trap is unmistakable:

• Cost basis per bitcoin: $57,686,457K ÷ 762,099 = $75,694
• Fair value per bitcoin: $51,649,675K ÷ 762,099 = $67,773

The second number is the real one. Bitcoin closed the quarter at $67,773 on Coinbase, which Note 3 identifies as the company's principal market for fair value purposes under ASC 820. The first number is an average of purchases made across years, most of them below the current price and a large tranche of them above it.

Q1 2026 is the useful teaching case precisely because cost basis sits above fair value. Anyone who grabbed the larger number and treated it as a mark overstated the stack by $6.04 billion. In quarters where bitcoin is up, the same mistake runs the other direction and understates it, which is harder to notice because the error is flattering in the wrong way.

Read it this way: cost basis is history, fair value is the mark.

For the period-end bitcoin price itself, don't compute it. Cite it. Strategy states it in the MD&A KPI table: $67,773 at 3/31/2026. That is the figure to carry into the snapshot.

Post-period holdings

Note 14, Subsequent Events, carries the position as of a later date: 818,334 bitcoins as of April 26, 2026, with a stated market price of $78,258 as of 4:00 p.m. Eastern that day.

This is genuinely useful and genuinely hazardous. It is useful because it tells you where the stack sits closer to the filing date. It is hazardous because it is a different as-of date than every balance sheet figure in the document. A CEBE snapshot built from 818,334 BTC and March 31 debt is not a snapshot of anything. Hold the whole capital structure at one date or don't build the snapshot.

One more line in Note 3 deserves attention: execution partners may extend short-term trade credits to the company and to MacroStrategy to buy bitcoin ahead of settling cash in their trading accounts. As of March 31, 2026, neither the company nor MacroStrategy had any outstanding trade credits payable, so there is nothing to add to senior claims this quarter.

Read the mechanic anyway, because it is the reason the line matters: while trade credits are outstanding, certain bitcoin is subject to a first-priority security interest and lien securing what is owed, which means part of the stack can be encumbered ahead of every other claimholder without any liability appearing on the balance sheet. A zero balance at one date is a fact about that date, not a feature of the structure. Check it every quarter.

Step 2Debt

Where: two places, and they disagree on purpose.

The balance sheet shows debt at carrying value:

Note 5 shows the same debt at principal:

Note 5 also discloses other long-term secured debt separately: aggregate net carrying value of $39.9 million and aggregate outstanding principal of $40.2 million as of March 31, 2026.

The maturities table at the end of Note 5 totals the whole structure at face:

Why CEBE uses face

The balance sheet number is $57.3 million lower than face. That gap is unamortized issuance costs: underwriting fees paid at issuance, capitalized and amortized into interest expense over the note's life. It is an accounting artifact of how the cost of raising the money gets spread over time.

But issuance costs do not reduce what has to be repaid. At maturity, the holder of the 2029 notes is owed $3,000,000K, not $2,984,138K. The senior claim on the bitcoin is the amount that must actually be satisfied, and that amount is principal.

The gap closes on its own as the notes approach maturity, with carrying value accreting toward face, which means anyone tracking debt at carrying value records a slow, steady increase in claims that never happened. The claim was always face. Only the bookkeeping was moving.

Fair value is a third number and it is not a claim at all. Note 5 marks the converts at $7,688,389K, well below both principal and carrying value. That mark reflects what the notes trade for, which moves with MSTR's share price, credit spreads, and rate expectations. It is disclosure, not obligation. A company cannot retire its debt by pointing at a quote.

One caution on the maturities table

Read the caption. Note 5 states that convertible note principal is placed in the table as if holders exercised their right to require repurchase on the respective Date of Holder Put Option, not at stated maturity. That's why the 2031 notes' $603,659K appears in the 2029 row and the 2032 notes' $800,000K in the 2030 row.

For CEBE this changes nothing: face is face regardless of which year bucket it sits in, and the total is the number you want. For anything involving timing, such as refinancing pressure or near-term claim maturity walls, the buckets mean put dates, not maturities. Don't reuse this table for duration work without re-reading the caption.

Step 3Preferred

Where: the mezzanine equity section of the balance sheet, and Note 9.

Preferred is the input that most often gets missed entirely, because of where it sits. It is not in liabilities. It is not in stockholders' equity. It occupies a band between them, called mezzanine equity, and a reader scanning the liability side for claims will scroll straight past it.

The balance sheet line, in full, is where the answer hides:

Two numbers on one line, $1.02 billion apart.

Why the gap exists, and why CEBE takes the larger number

The 10-Q explains it directly in the significant accounting policies discussion: carrying values are initially recognized based on proceeds received net of issuance costs, and are not accreted toward redemption value unless redemption becomes probable.

So the carrying value is the cash that came in the door, minus fees, frozen at issuance. Strategy has issued preferred continuously at prices around and below the $100 stated amount, and issuance costs come off the top. The liquidation preference is what the preferred is owed ahead of common in a wind-up. Since CEBE asks what the common actually has a claim on after everyone senior is satisfied, the liquidation preference is the number.

Note that carrying value here is sticky in the wrong direction: it does not accrete unless redemption becomes probable, which means it can sit a billion dollars below the real claim indefinitely. Unlike the debt gap in Step 2, this one does not close on its own.

The series detail

Note 9's key terms table breaks the aggregate into its parts:

The series total ties to the balance sheet aggregate exactly. Always run that check; it catches transcription errors immediately.

Two things to carry forward. STRE is euro-denominated, with a €100 stated amount translated to $891,560K at the filing's rate. That's an FX flag on the claim, not a change to how it scores. And STRK's preference floats daily. Per the July 2025 amendment to the Certificate of Designations, it is the greatest of the $100 stated amount, the last reported sale price, or the 10-day average close. That is a daily floating maximum with a $100 floor, recalculated each day rather than stepped permanently upward. It is not a ratchet: a preference that rose with STRK and stayed there would be a different instrument, and the distinction matters because a floating maximum falls back as readily as it rises.

The float is also narrower in scope than the mechanic alone suggests. It governs liquidation entitlement, issuer-elective redemption, and the Section 6(a)(i) payment provision. Fundamental-change repurchase and conversion fall outside it. Note 9's additional-information disclosure references the $100 per share preference and the ten-trading-day mechanic; the aggregate stated in the filing is computed on the stated amount.

Notional versus accrued

The $10,004,676K is notional: the principal-equivalent claim. Accrued but unpaid dividends are a separate, smaller, current obligation and they appear on the liability side of the balance sheet:

Don't double-count it inside the notional, and don't ignore that it exists. Note 10 shows the flow-through cost: $229,527K of preferred dividends charged against the numerator in the quarter's loss-per-share computation. That is the running cost of the wrapper, and it is one of the clearest reasons preferred cannot be treated as equity just because it sits in the equity half of the balance sheet.

Step 4Cash

Where: top of the balance sheet. The easiest of the five.

Cash nets against senior claims because it can be applied to them. Every dollar of cash on the balance sheet is a dollar of debt or preferred that does not have to be satisfied out of the bitcoin.

What does not net: receivables, prepaid expenses, deposits, or anything else in current assets. Those are operating working capital. Accounts receivable of $122,257K is money owed by software customers on ordinary terms; treating it as a claim offset assumes it converts to cash on demand and at par, and it does neither.

Restricted cash is presented on its own line for a reason. It is committed, typically to collateral or escrow. At $2,026K here it changes nothing (thirty bitcoins at the period-end price), but the convention matters more at companies where restricted balances are large, and it should be applied the same way everywhere rather than case by case.

Step 5Shares

This is the input that goes wrong most often, because the 10-Q publishes five different share counts and four of them are wrong for CEBE. Every one is real, correctly labeled, and correct for its own purpose. That is what makes this the hard step: nothing here is an error in the filing.

The instruction

Use the balance sheet parenthetical. Period-end, both classes, GAAP issued and outstanding.

The statement of stockholders' equity carries the same figures as its ending balances, so the two tie.

One limitation of the source, stated plainly: the parenthetical is presented in thousands, so this count is exact only to the nearest thousand shares. The 10-Q publishes no unrounded period-end count anywhere. The exact-to-the-share figure exists only at the cover date, which is the wrong date. That is a property of the filing, not a lookup failure, and it is not worth hunting for.

The rest of this step is about the four counts you will be tempted by instead.

The mistake worth making slowly

The most damaging denominator error in this filing is not a trap. Nobody is hiding it. It is in the footnote whose entire job is to divide by a share count, it is labeled "Basic," and reaching for it is a reasonable thing an intelligent reader does on the way to being wrong.

Note 10, the EPS footnote: weighted average.

Weighted average is 333,913K against period-end 345,926K, 12.0 million shares lower, a 3.6% gap. That gap is the arithmetic signature of a heavy issuance quarter. The statement of stockholders' equity shows 33,469K Class A shares issued under public offerings during Q1 plus 277K on option exercises; shares issued in March counted for only a few days of the average, and shares issued on March 31 counted for essentially none.

Weighted average is correct for EPS. Earnings accumulate across a period, so the denominator has to accumulate the same way, since charging a full quarter's loss against shares that existed for six days would overstate the loss per share. The convention exists for a good reason and it is the right convention there.

It is the wrong convention here. CEBE is not a flow measure. It is a claim on assets at a point in time, and every share outstanding on March 31 has an equal claim on the bitcoin held on March 31, regardless of whether it was issued in January or the morning of the 31st. A share does not get a partial claim on the treasury for having arrived late.

Using weighted average reports CEBE at 157,304 sats instead of 151,842, which is 3.6% high.

Three things about that error are worth sitting with.

It is the largest of the denominator errors available in this filing, larger than the cover-page date mismatch, larger than the same-date pair discussed below.

It runs in the flattering direction. It makes the quarter look better. Errors that make the number worse get caught, because someone always checks a disappointing result. Errors that make the number better get published.

And it grows with issuance. The heavier the ATM quarter, the wider the gap between weighted-average and period-end, which means this error is largest exactly in the quarters that matter most, and nearly invisible in quiet quarters where you might first calibrate against it. In an annual filing, measured across four quarters of issuance, it is wider still.

If you take one thing from this step: the word "Basic" in a filing does not mean "the shares that exist." It means "basic as opposed to diluted." Those are different claims.

The cover page: right count, wrong date

Bottom of the first page, above the table of contents:

This is the most precise count in the filing, exact to the share with no rounding, and precision is exactly what makes it tempting. But it is as of a date after the quarter ended, because the cover reports the count near the filing date. Pairing 350,447,872 cover-page shares with March 31 bitcoin folds 26 days of ATM issuance into a period-end snapshot and understates CEBE by 1.3%.

The cover count has one excellent use. Note 14's post-period bitcoin figure carries the same as-of date, April 26, 2026. Cover shares and Note 14 bitcoin are date-consistent with each other, so together they build a coherent post-period view. They simply cannot be mixed with anything dated March 31.

One line on the issuer's own headline basis

There is a fifth count, and it is the denominator behind the number Strategy leads with: Assumed Diluted Shares Outstanding, 378,834,000 at 3/31/2026 (MD&A KPI table), meaning basic plus every convertible note, convertible preferred, and equity incentive award assumed converted, which is what BPS divides by, so any BPS figure and any CEBE figure are separated by a denominator gap of 33 million shares before either metric's treatment of claims is even considered.

The smell test: two counts, one date, one label

Close the step with the pair that should make you check everything else.

Same label as the balance sheet. Same date as the balance sheet. Different number: Class A of 326,582K against the parenthetical's 326,286K, 296 thousand shares apart.

Footnote (1) explains it without apology: the figure includes shares sold but not yet settled as of March 31, 2026. Contracted, not yet issued: outside the GAAP outstanding count, inside the company's KPI count. Both definitions are defensible. Neither is hidden.

The CEBE effect is trivial, 0.086%, roughly 130 sats. Which is precisely why it belongs at the end rather than the middle: it is not a material error, it is a calibration check. Two numbers carrying the same label on the same date in the same document differ by construction, and the only reason anyone knows that is that someone read footnote (1).

If a figure that small can diverge in a filing this carefully assembled, then "it says Basic and it's dated March 31" was never sufficient grounds for using a number. Pick the definition, write it down next to the figure, and apply it every quarter, or the series stops being comparable to itself, which is the one failure a tracker cannot recover from.

The anti-dilution convention in loss periods

Look again at Note 10: basic and diluted weighted-average shares are identical, both 333,913K. Basic and diluted loss per share are also identical, both $(38.25).

That is not a typo and it is not a company with no dilution. It is the anti-dilution rule. In a period with a net loss, including potentially dilutive securities would reduce the loss per share, which makes the result look better than reality. GAAP forbids it. So in loss periods every convertible instrument, option, and unvested unit is excluded, and diluted collapses onto basic.

Note 10's companion table shows exactly what got excluded, 32,433K weighted potential Class A shares:

Two consequences worth holding onto.

First, the diluted share count is not a usable time series across profitability changes. A company that swings from profit to loss will show diluted shares collapsing to basic, a mechanical artifact that looks like dilution reversing. Nothing reversed. Thirty-two million shares of overhang are still there; the convention just stopped displaying them.

Second, this is a further argument for CEBE's period-end basic denominator. Basic is a fact about shares that exist. Diluted is a calculation whose definition changes with the sign of net income, and any metric built on it inherits that instability. Convertible overhang is a real thing worth tracking. It just belongs in its own analysis, not baked into the denominator of a point-in-time claim measure.

Aside: the authorized-shares parenthetical

Sitting in the same parenthetical is a number that is not an input to anything: 10,330,000,000 Class A shares authorized against 326,286,000 issued.

Authorized shares are capacity, not dilution. No claim, no economics, nothing in the CEBE numerator or denominator. Roughly 3% of the authorization is used.

It's worth a glance anyway, because capacity constrains strategy. A company approaching its authorized ceiling has to go to shareholders for an increase before it can keep issuing equity, and that's a scheduled, visible event with a timeline. Strategy has no such constraint at 3% utilization, so the runway is effectively unlimited for planning purposes. The same parenthetical shows preferred authorization split between the designated Series A (424,953K authorized, 98,881K issued) and undesignated preferred (580,047K authorized, none issued), which is capacity to create new series without a charter amendment.

Read it as a governance signal about what the company can do next. Never as a number that belongs in the calculation.

Step 6Assemble the Snapshot

Five inputs, all at March 31, 2026, all from this filing:

What the decoys would have produced

Same filing, same quarter, each error taken one at a time:

Every one of these is reachable by reading a real number off a real page of the actual filing. The two largest both run in the flattering direction, and both land within a few percent of correct, which is exactly what makes them dangerous. A number that is obviously wrong gets caught. A number that is quietly wrong, and pointed the right way, gets published.

For contrast, the MD&A reports bitcoin per basic share at 220,118 sats for the same date. That figure is gross: it divides all 762,099 bitcoins by basic shares and deducts nothing. CEBE says 151,842. The 68,276-sat difference is not a disagreement about arithmetic. It is the $16.05 billion of debt and preferred standing between the common shareholder and the stack, which one measure subtracts and the other does not.

And BPS itself divides by Assumed Diluted Shares Outstanding (378,834K), not basic, so a BPS-to-CEBE comparison differs in both numerator and denominator at once. Know which count any published figure uses before setting it beside anything.

Annual FilingsWhat the 10-K Adds

The annual report carries the same five inputs in the same places, with the same traps. Read it the same way, with four differences:

Audited, with an opinion. The 10-K includes the auditor's report and any critical audit matters. Digital asset existence and custody arrangements are frequently a CAM at treasury companies, and it's worth reading. It tells you what the auditor thought was hard.

Full-year weighted average. The gap between weighted-average and period-end shares is typically much wider over four quarters of issuance than over one, which makes the Step 5 denominator error correspondingly larger in annual filings.

Complete debt terms. The 10-K debt note carries the full terms of each instrument, including conversion prices, put dates, redemption provisions, and covenants, where 10-Q notes often abbreviate and cross-reference the prior 10-K.

Risk factors are current. Item 1A is fully restated annually; 10-Qs only report changes. For custody arrangements, counterparty concentration, and covenant risk, the 10-K is the source.

Everything else is identical: mezzanine placement, face versus carrying, the cover-page date gap, anti-dilution in loss periods.

For Other CompaniesSame Five Inputs, Different Filing

The five inputs are universal. The document isn't.

US filers (SEC EDGAR): 10-Q quarterly, 10-K annual, 8-K for material events. Strategy, Strive, ProCap, Nakamoto, DDC. Same Archives pattern, same note structure, same traps.

Japan (Metaplanet): Tanshin quarterly summaries and securities reports via the TSE and the IR page. Preferred and convertible terms in Japanese-language filings; JPY-denominated claims carry an FX flag.

UK (Smarter Web): RNS announcements and interim/annual reports. Convertible loan note terms in admission documents rather than a periodic filing footnote.

Sweden (H100 Group): NGM Nordic SME interim reports.

France (Capital B): Euronext filings; OCA convertible terms in AMF-registered prospectuses.

Brazil (Méliuz, OranjeBTC): CVM filings: ITR quarterly, DFP annual. Debenture and warrant terms in the offering documentation, often in a separate document from the financials.

In every jurisdiction the five questions are the same: How much bitcoin? What debt, at face? What preferred, at liquidation preference? How much cash? How many basic shares at period end?

And in every jurisdiction the same discipline applies: find each number's location, write the location down next to the number, and never carry a figure forward that you cannot point back to a specific note or line.