Skip to content

Date a page whose claim is about a version

Draft - this is a knowledge-management practice

Unfinished. It is here because the gap is worth showing, not because it is ready to be read as settled.

How knowledge gets organised, independently of any tool that stores it.

"Treat warnings as errors" was true in 2019 and will be true in 2035. "The resolver refuses this upgrade unless both settings are present" was true of one version of one tool, and something changed it three releases later. Both are worth writing down. Only the second has an expiry date, and a knowledge base that stores them identically is lying about one of them.

The rule: when a page's claim is bound to a version, make the binding visible on the page, in its name, and in the order it is listed.

Three places it shows

  • A date in the filename. 20260911_dependabot_declines_transitive_security_updates.md says when the claim was made before anything is opened. A page arrived at from search is read out of context, and the date is the context that matters most.
  • An explicit expiry, such as OKF's stale_after. This is the part that turns a human worry into a mechanical question: without it, "is this still true?" is asked by whoever happens to wonder, which is nobody. With it, a linter asks.
  • Newest first. Version-bound pages are listed with the recent ones at the top, because recency is a proxy for still being true. Idea-bound pages are ordered by how the argument builds, because their age is beside the point.

Why not date everything

Uniformity is tempting and wrong here. A date on a principle invites a reader to discount it for being old, which is precisely the wrong inference: an idea that has survived six years is better evidence than one written last week. Dating it imports a signal that does not apply and reverses its meaning.

So the date is not metadata about the page. It is part of the claim, and it belongs only on pages whose claim includes it.

What this is not

It is not updated:. A last-touched timestamp records activity on the file and says nothing about whether the world still matches it; a page can be edited for typography and remain wrong. The date here records when the claim was observed to hold, which is a statement about the subject rather than the document.

Nor is it a substitute for checking. An expiry date does not make a page correct until it passes, and it does not make it wrong afterwards. It marks the moment the claim stops being load-bearing without evidence, which is all a date can honestly do.

The following pages link to this page: