The Same Three Months Were a Loss of $2.49 a Share and a Profit of $0.66. Both Numbers Are Right.

Two numbers, one quarter, no contradiction

For the three months ended December 31, 2025, Coinbase lost $2.49 a share.

For the same three months, Coinbase earned $0.66 a share.

Neither figure is a mistake, a revision or a typo. They are the same business over the same ninety-two days, measured two ways that both have a legitimate claim to being the number. The first is GAAP diluted earnings per share — what the accounting rules produce. The second is the adjusted figure, which is what the company led with.

A swing from a loss to a profit is the extreme case, and it is not a rare one. What follows is what the two numbers are, how far apart they get, which direction the gap runs, and — the part almost nothing tells you — how to know which one you are holding.

What the two numbers actually are

GAAP diluted EPS = net income attributable to common shareholders ÷ weighted-average diluted shares, computed under Generally Accepted Accounting Principles. It is audited, it is in the filing, and every company computes it the same way.

Adjusted (or "non-GAAP") EPS = the same figure with specified items removed — typically stock-based compensation, acquisition-related amortisation, restructuring charges and one-off gains or losses. Each company defines its own adjustments and must reconcile them to GAAP in the release.

The second is not a lesser number. It is an argument: these items are in our results but they do not describe the business you are buying. Sometimes that argument is excellent — a decade of amortisation from an acquisition that closed years ago tells you very little about this quarter's operations. Sometimes it is self-serving. It is always the company's own view.

What it is never is comparable across companies by default, because two companies adjusting for different things have not computed the same measure.

The scale of it

Marvell Technology, for the quarter ended May 2, 2026:

Twenty times. $0.80 ÷ $0.04 = 20. Marvell carries heavy acquisition-related amortisation from the Cavium and Inphi deals, and that charge is real cash spent, in the past, on assets it still operates. The GAAP number counts it against this quarter. The adjusted number does not. Both statements about Marvell are true and they differ by a factor of twenty.

That is what a headline like "beat by four cents" is often measuring — four cents against a base that may be $0.04 or $0.80 depending on a choice nobody stated.

Now the part that surprises people

The reflex is that "adjusted" means "flattered": add back the ugly items and the number goes up. On most of the companies here it does. But it runs the other way often enough that the reflex is a bad rule.

Occidental Petroleum, quarter ended March 31, 2026:

The GAAP figure is roughly three times the adjusted one — $3.13 ÷ $1.06 = 2.95. Dell runs the same direction over the quarter ended May 1, 2026: $5.24 GAAP against $4.86 adjusted.

The mechanism is the mirror image of Marvell's. A one-off gain — an asset sale, a divestiture, a legal settlement in your favour — lands inside GAAP net income and inflates it. Management strips it out precisely because it does not recur, and the adjusted number comes out lower. A company doing this is being conservative about its own results, which is the opposite of the motive the cynical reading assumes.

So the honest formulation is not "adjusted is the flattering one." It is:

Adjusted EPS is management's estimate of what repeats. Whether that is higher or lower than GAAP depends entirely on whether the unusual items in the quarter were costs or gains.

Hold that and both directions make sense. Hold "adjusted means flattered" and Occidental looks like an error.

Six quarters, side by side

Each row is one quarter of one company, with both figures taken from a single source that printed both:

Two of six run against the reflex. A reader who took away only "adjusted is bigger" would be wrong on a third of this sample.

Where you will meet this without being told

This site is a working example, and it is worth being specific rather than abstract about it.

The quarterly earnings series behind every chart here is GAAP diluted EPS. That is a deliberate convention: it is the one basis every company computes identically, so it is the only one on which a chart of sixty-six companies means anything. We checked it rather than assumed it — every tracked filer's stored series was compared against the diluted-EPS figures in its own SEC filings, and fifty agree on every quarter tested.

Our written coverage, meanwhile, quotes what the company led with in its release, which is usually the adjusted figure. That is also right: an article about a print should report the number the print was about.

The consequence is that on nineteen of the companies tracked here, an article quotes a quarterly EPS figure that appears nowhere on the chart beside it. Not because either is wrong — because they are different measures, and until this paragraph nothing said so.

If that can happen on one site with one convention, deliberately chosen and verified, it happens constantly across the sources a reader actually uses, where no convention is stated at all.

We are not immune to the confusion either. Until days ago the Marvell series on this site stored the adjusted figure rather than the GAAP one — the chart showed $0.80 for the quarter above. It was corrected, which is why the row now reads $0.04. Of all the illustrations available, our own is the most honest one: the distinction is easy to lose even when you have written the convention down.

What is deliberately not in the table

Three exclusions, because each would make the picture look worse than it is:

• Full-year guidance figures are excluded. Sixteen candidate rows were annual guides, not quarterly results. Setting a full-year guide beside a single quarter's chart point manufactures a difference that means nothing at all.
• Stock splits are not basis differences. Three companies' stored figures differ from their as-filed ones by exactly the split ratio — 10×, 1/5, and about 1/4. The stored series is split-adjusted and correct; nothing about GAAP or adjusted is involved, and none of the three is counted as evidence here.
• A handful of tickers run the other way round, with the adjusted figure in the series rather than GAAP. They are excluded from every count above, because including them would assert the reverse of what this piece argues, and their repair is its own piece of work.

What you can check here, and what you cannot

This matters more than usual on a piece whose whole claim is that two numbers both check out, so it is worth stating exactly where each side comes from.

The "GAAP diluted" column is checkable on this site. Every figure in the table is the value stored in the company's own series and plotted on its chart page; you can open any of them and read it off.

The verification against the filings is not. This repository stores an index of SEC filings — form, filing date, period covered, accession number — and no dollar figures. The comparison that established the convention was run against the filings themselves, which live at the SEC and not here. So a reader can confirm what we store and must go to the filing to confirm it is what the company reported. That is a real limit and it is the reason this section exists rather than a single word "verified" covering both halves.

And the anchor of the whole piece is the weakest-verified row in it. Coinbase's figure could not be confirmed against a quarterly filing, because its fiscal fourth quarter has no separate quarterly context in the filing data — the annual covers it. Deriving it as the full year less the first nine months gives −$2.40 against the −$2.49 stored, and that nine-cent residual is larger than rounding.

The likely explanation is instructive in its own right: diluted EPS is not additive across periods. The diluted share count changes when a company swings between profit and loss, because potentially dilutive shares are excluded from a loss-making period and included in a profitable one. So a fourth quarter genuinely is not the year minus nine months, and the two figures disagreeing by nine cents is expected behaviour rather than a discrepancy to resolve.

What supports the −$2.49 is two independent sources inside this site — the stored series and our own published coverage of that quarter — agreeing. That is decent provenance. It is not the filing, and calling it "verified" alongside rows that are would be exactly the unlabelled conflation this piece is about.

What to do with this

Three habits, and they cost nothing:

1. When you meet an EPS figure, find out which one it is before comparing it to anything. A beat or miss is meaningless without it, because the consensus estimate has a basis too, and it is usually the adjusted one.
2. Never compare across the line. A GAAP figure for one company against an adjusted figure for another is not a comparison. This is the single most common way the two get mixed, because each source is internally consistent and the reader is combining them.
3. When they diverge a lot, read the reconciliation. Every release that quotes an adjusted figure must show its bridge to GAAP. That table is where the actual information is: it names each excluded item and its size, and it will tell you within a minute whether a 20× gap is a decade-old amortisation schedule or something you should care about.

The gap between the two numbers is not a scandal and it is usually not even a disagreement. It is two questions — what happened and what recurs — with two different right answers. The only real error is not knowing which one you were handed.

The GAAP figures for all six companies are the values stored in the EPS series of every company we track and plotted on each company's EPS chart page here; they can be read off the site directly. The adjusted figures are the ones the companies led with, as quoted in this site's own coverage of each print — every row is a quarter for which a single article printed both bases, so no cross-source matching of periods was required. The convention that the stored series is GAAP diluted was established by comparing every tracked filer's series against the diluted-EPS figures in its SEC filings: fifty agree on every quarter tested, nine differ for three separate and identified reasons (an adjusted basis, a stock split, and one scaling fault), and seven have no comparable quarterly figure to test against. That comparison is not reproducible from this repository — the SEC data stored here is a filings index carrying no dollar amounts — so it is cited as method rather than offered as something a reader can re-run without going to the filings. The Coinbase row is the exception noted above: corroborated by two sources within this site and not confirmed against a quarterly filing figure, for the reason given. The nineteen-company count excludes full-year guidance figures, which would otherwise have added sixteen false matches. Multiples are stated with both operands so the division can be checked. No share price, market-capitalisation or share-move figure appears above.