The formats that will still open in 2125

Media failure is the risk everybody plans for: disks die, so you keep copies. Format obsolescence is the one that actually eats collections, and it is harder to see because the files remain perfectly intact right up until nothing can open them.

What makes a format survivable

An open, published specification. Written down, publicly available, complete enough that somebody could implement a reader from it alone. This is the single most important property and the one most often traded away for convenience.

Multiple independent implementations. If one organisation’s code is the only thing that can read a format, the format’s lifespan is that organisation’s lifespan. Several implementations, from different sources, means the knowledge is distributed.

Wide adoption outside your use. Formats survive because many people depend on them. A technically excellent format with three users is a worse preservation bet than a mediocre one everybody uses.

Simplicity. Every feature is a thing a future reader has to implement. Layered, extensible, DRM-capable formats have more places to fail and more reasons a future implementation will only partially work.

No dependency on external state. A format that references a server, a licence check or a font nobody has is not self-contained, and self-contained is the whole game.

The awkward part about what that leaves

Applying those criteria honestly leaves a shorter list than most people want: plain text and lightly marked-up text, simple raster formats with published specifications, a small number of widely implemented audio and video encodings, tabular data as delimited text rather than spreadsheets.

Things that fail the test include most office document formats, most design and editing project files, anything with rights management, and anything whose specification lives inside a product rather than in a standards body.

That is a real loss of fidelity, and pretending otherwise would be dishonest. A document flattened to text loses layout; a project file flattened to its output loses editability. The trade is deliberate: less fidelity now in exchange for the material still opening later, and it is a trade worth making explicitly rather than discovering by accident.

Migration is the actual practice

Choosing good formats is the beginning. The practice that keeps an archive alive is periodic review — checking whether the formats in it are still widely readable, and moving material forward before the answer becomes no.

This means an archive is not a thing you finish. It is a maintained thing, on a schedule, with somebody responsible for it, forever. Every honest preservation cost model is a recurring cost model, and any offer that takes a single payment for permanent storage is either subsidising your material from other revenue or quietly planning not to be there.

The test that cuts through all of it

Could somebody open this with no account, no licence, no internet connection and no software from the company that made it?

If the answer is no, it is not preserved. It is hosted, which is a different service with a different lifespan.