XBRL

How it works

A filing marked up in XBRL is not a spreadsheet of labelled numbers. It is a set of facts, and a fact has four parts: an element from a taxonomy (the concept — what the number is), a context (which entity and which instant or period it covers), a unit, and optionally one or more dimensions that narrow it further. "Revenue" is not a number. It is a concept against which a filer has reported several facts — a year, four quarters, a restated prior year, and a dimensional breakdown by product line or geography — and pulling "the" revenue out means choosing among them.

The markup used to travel beside the document. Under the current regime it travels inside it: inline XBRL embeds the tags in the same HTML file a person reads, so there is one artefact rather than a human-readable document plus a machine-readable exhibit that could quietly disagree. The SEC's staff interpretations record that all relevant phase-in periods for operating company financial statements and fund prospectuses ended in September 2021. Europe went the same way for a different reason: under the European Single Electronic Format, listed issuers tag IFRS consolidated statements in inline XBRL against a taxonomy ESMA updates annually, most recently with the 2025 IFRS taxonomy.

Where standardisation stops

Extension elements. A filer that cannot find a suitable element in the standard taxonomy creates its own. That is permitted and sometimes correct, and it is also where comparability goes. The SEC's sample comment letter on XBRL disclosures asks companies why they used a custom tag where an existing US GAAP element would have served, and asks separately why the element used for the same line item changed between periods. Both questions exist because both happen.

Dimensions. The same staff guidance tells filers to use the predefined tables and axes as they exist in the US GAAP taxonomy, and says creating new hypercubes and dimensions should generally be avoided. The advice is not always taken, and a consumer who ignores dimensions altogether will sum a total together with the slices that already compose it.

Mechanical errors. The staff letter also notes shares outstanding tagged with materially different values on the cover page and in the balance sheet because the two were scaled differently. A tagged number is not a verified number.

What the SEC's own layers flatten

Two free surfaces read the tags for you, and each gives something up. The XBRL APIs return facts by company or by concept, and the frames endpoint assembles one concept across filers by taking the fact last filed that most closely fits the period — a mix of vintages rather than a snapshot, which is the point-in-time problem wearing a convenient interface. The Financial Statement Data Sets go the other way: published per filing and presented as filed, with no standardisation applied, which makes them usable for reconstructing what was reported and useless as a comparable panel straight out of the box. How to get SEC filings as data covers both endpoints and their limits.

Why it matters here

The practical consequence is the one that costs money: two vendors reading the same filing legitimately produce different fields. Every step between the markup and a column in an API response is a judgement — which element counts as the headline figure, whether a company's extension maps onto a standard line or is dropped, which dimensional slice is "the" segment, what to do when the filer changed elements in the third year of the series. None of that is visible in the response, and none of it is wrong.

So the question to ask a fundamentals vendor is not whether it uses XBRL. It is what happens to the parts XBRL does not settle. SEC EDGAR hands you the facts and no opinion, free and keyless, and the assembly is yours. sec-api.io sells the conversion and the section extraction. SEC EDGAR MCP puts the same free endpoints behind an assistant. SimFin and the standardising vendors sell the mapping itself, which is the expensive part and the part you cannot audit. Daloopa exists because of what is never tagged at all — segment splits, operating KPIs and guidance that live in a press release or on slide 16 — and links each figure back to the page it came from, which is the honest answer to an untagged number.

A single field is rarely worth an argument. A model built on fifteen of them, sourced from two vendors, is where this term stops being vocabulary. Fundamental data and research platforms is the shelf, and what is genuinely free is where the free half of it stops.