Non-GAAP

How it works

Generally accepted accounting principles produce a number. A company may also publish a different number, arrived at by excluding or including amounts the accounting measure does not, and that second number is a non-GAAP financial measure. The SEC's own definition, in the release adopting Regulation G, is a numerical measure of a registrant's historical or future financial performance, financial position or cash flows that either excludes amounts included in the most directly comparable measure calculated under GAAP, or includes amounts excluded from it.

The adjustments come from the company. Typical ones are share-based compensation, restructuring charges, acquisition and integration costs, amortisation of acquired intangibles, impairments, legal settlements and currency movements. Nothing in the definition constrains the list, which is the point worth holding on to: "adjusted EPS" is not a measure with a specification, it is a measure with an author.

What the rules do constrain is the disclosure. Regulation G requires a company that publicly discloses a non-GAAP measure to present the most directly comparable GAAP measure alongside it and to give a quantitative reconciliation between the two. Item 10(e) of Regulation S-K adds requirements for figures inside SEC filings, including that the comparable GAAP measure be presented with equal or greater prominence, and it prohibits several specific presentations — putting non-GAAP measures on the face of the financial statements or in the notes, and using titles confusingly similar to those of GAAP measures among them. The staff interpretations go further on substance: adjustments that change the recognition and measurement principles GAAP requires are treated as individually tailored accounting principles and may make the presentation misleading.

So the regime is not "GAAP only". It is: publish what you like, show the comparable statutory figure next to it, and show the arithmetic that gets from one to the other.

Where it breaks in a data field

That reconciliation is a disclosure obligation, not a data format. It is a table in a filing or an earnings release, and the two numbers it bridges leave the company through different doors.

The as-reported figure is in the financial statements, tagged, and therefore in the SEC's structured surfaces — the XBRL facts behind the EDGAR APIs, and the Financial Statement Data Sets, which are published per filing and presented as filed with no standardisation. The adjusted figure usually appears first in a press release, often as a headline and a bridge table, and much of that material is not tagged as the statements are. Two numbers, two provenances, one English word.

A vendor sitting between you and both of them has to pick. Some serve the as-reported figure. Some serve the company's adjusted figure, because that is what a sell-side model uses and what estimates are quoted against. Some serve their own standardised measure, which is a third thing again: neither the statutory figure nor the company's, but the vendor's normalisation applied consistently across its universe. All three can be labelled eps. See XBRL for why even the as-reported side of that is less settled than it sounds.

Why it matters here

The failure is quiet and it compounds. Joining one vendor's adjusted figure to another's reported figure produces a ratio, a growth rate or a screen result that looks ordinary and describes nothing — and unlike a missing value, it does not announce itself. Ranking a universe on a field that is adjusted for some rows and reported for others is the same error at scale.

Three questions are worth asking any fundamentals source on this site before it becomes a dependency. Which basis is each field? Is the basis the same for every company and every period in the series? And is the other basis available, so a check is possible at all? A source that cannot answer the first question has not decided, which means the answer varies by row.

Where the cards fall is fairly consistent with how they are sold. TIKR and Fiscal.ai sit on standardised institutional data, where the point is cross-company consistency rather than fidelity to one filing. Daloopa works the other end, extracting the bridge itself — the adjustments, the KPIs and the guidance — with each figure linked back to the page it was taken from, which is the only form in which an adjusted number is auditable. Financial Datasets and Wisesheets pull from the filings, which puts them nearer the as-reported side. SEC EDGAR is the statutory figure and nothing else, free, and how to get SEC filings as data is the route to it.

Whether an adjusted figure is a better description of a business is a question for the reader of the filing. Which one is sitting in the column is a question about the data, and it has an answer. Fundamental data and research platforms is where the rest of that shelf is, and point-in-time is the sibling question: not which basis, but which vintage.