When a public administration is told its documents are stored in “an ISO standard format,” the assumption is reasonable: an ISO standard ought to be a clean, implementable specification that any qualified software vendor can support. Standards exist precisely so that nobody is locked to a single supplier.
OOXML — ISO/IEC 29500, the format behind Microsoft’s docx, xlsx and pptx files — does not work this way.
The standard is split into two conformance classes. Strict is the clean version: a modern document format, free of legacy baggage, that an independent implementer could reasonably support. Transitional is everything else: a vast catalogue of compatibility features, deprecated elements, platform-specific behaviours, and references to undocumented quirks of Microsoft Office versions from the 1990s. The Transitional class exists to ensure that documents converted from the old binary doc, xls and ppt formats can be represented in XML without loss.
There is one detail that matters above all others: Microsoft Office has never produced Strict OOXML by default. The option to save in Strict format is available in the installed desktop applications but is absent from the browser-based versions of Microsoft 365 — and Microsoft’s various editions have long differed in which features they offer, with the macOS version historically providing a different set of options from the Windows version. The “ISO standard” that public administrations are actually storing their documents in, when they use Office, is Transitional — the messy one. Strict is a feature you can find if you know where to look, on the platforms where Microsoft has chosen to support it. That is not the treatment a serious open standard receives.
This has consequences that go well beyond a technicality.
The standard codifies undocumented legacy behaviour. Transitional OOXML contains compatibility flags whose specification amounts to “behave like Word 95” or “lay out footnotes like Word 97.” These are not formal definitions. They are references to the behaviour of specific commercial software products released more than thirty years ago — products whose layout algorithms were never published. An independent implementer wishing to render such a document correctly must reverse-engineer software from the Windows 95 era. This is not standardisation in any meaningful sense; it is the codification of one vendor’s implementation history as a global norm.
The standard perpetuates known bugs. Excel famously treats 1900 as a leap year — it was not — because Lotus 1-2-3 did so in the 1980s, and Microsoft chose binary compatibility with Lotus over arithmetic correctness [1]. OOXML Transitional preserves this bug. The default workbook setting in every xlsx file you have ever opened encodes a date arithmetic error from the era of MS-DOS. A spreadsheet calculating durations across February 1900 will produce wrong answers, and the standard requires this.
The standard includes obsolete graphics formats. Vector Markup Language (VML) was submitted by Microsoft to the W3C in 1998 as a candidate vector graphics standard. The W3C rejected it in favour of SVG. VML should have died there. Instead, it lives on inside OOXML Transitional, because documents converted from doc files contain it, and Microsoft Office continues to emit it. Implementers must support both VML and its modern replacement, DrawingML, to handle real-world files.
The conformance class problem is structural. Strict was meant to be the future and Transitional the temporary bridge. Two decades after standardisation, Transitional remains what Office produces, what users receive, and what any competing implementation must support to be useful. The clean standard exists on paper. The standard that exists in practice — and that Microsoft Office produces by default — is the messy one.
For public administrations, this matters in three specific ways.
For archives. A document format that depends on undocumented behaviour of 1990s applications is not a safe long-term archival format. The ISO label provides a false reassurance: the parts of the standard your documents actually use are precisely the parts that are least specified and most dependent on a single vendor’s tooling.
For procurement. Specifying “ISO/IEC 29500” in a tender does not guarantee interoperability or vendor neutrality. It guarantees that documents will conform to a specification of which the practically deployed variant is, in effect, whatever Microsoft Office does. This is the opposite of what an open standard is meant to deliver.
For sovereignty. European institutions, national governments, and regional administrations increasingly recognise that the choice of document format is a sovereignty question. A format whose definitive reference implementation is a single American company’s commercial product cannot serve as the technical foundation of European digital autonomy — whatever its ISO number.
The alternative is not hypothetical. The OpenDocument Format (ODF), ratified as ISO/IEC 26300 twenty years ago this month, was designed from the outset as an implementer-neutral standard. Its specification is complete, self-contained, and does not require knowledge of any specific commercial product’s history. Multiple independent implementations exist. It is, in the proper sense of the term, an open standard.
For administrations weighing format policy, the question is not whether OOXML is “a standard.” It is. The question is what compliance with that standard actually entails, what it demands of implementers, and whether that serves the long-term interests of the institutions storing their work in it.
For those interested in the technical detail behind these claims, we attach a companion deep-dive [2] cataloguing the Transitional features, their categories, and the specific structural problems they introduce.
[1] The history of the 1900 leap year bug is well documented. Joel Spolsky, who worked on the Excel team at Microsoft in the early 1990s, recounted in My First BillG Review how Excel inherited the bug from Lotus 1-2-3 to preserve binary compatibility. Microsoft’s own support documentation openly acknowledges the bug and explains why it will not be fixed: doing so would invalidate every date in every existing Excel worksheet.
[2] The companion deep dive document in PDF format, cataloguing the Transitional features, their categories, and the specific structural problems they introduce: A Standard in Name Only a Deep Dive