A design-tool company in the mould of Canva stamps every page of its 1200-page internal documentation site with a last-verified date and marks anything older than 180 days as unverified. Support questions drop in the first quarter. What has the team given up and when does the bill arrive?

What is gained

A reader can price a page before trusting it. That is a real and underrated gain: the cost of stale documentation is not that it is wrong, it is that the reader cannot tell which parts are wrong, so they verify everything or nothing. A visible verification date turns a binary trust decision into a graded one, and it explains the early drop in support questions — readers stop escalating pages they can see are fresh.

It also makes decay measurable. Before the badge, "our docs are stale" is an opinion. After it, the percentage of the corpus inside the window is a number, and a number can be argued about with evidence.

What is paid

The grey count grows faster than anyone expects. With a 180-day window and 1200 pages, keeping the whole site green needs roughly 1200 / 180 ≈ 7 page verifications every single day, forever, with no new pages. Nobody staffs that, so within two quarters a large share of the site is grey.

Then the second-order effect arrives. Grey is read by leadership as a team failure, so re-dating becomes the cheapest way to make the number move — and re-dating a page costs one click while actually reading it against the system costs twenty minutes. The signal degrades into a record of who remembered to click. That is the bill: a metric that was honest in quarter one is a compliance ritual by quarter three, and now the organisation has false confidence rather than honest uncertainty, which is strictly worse than where it started.

The decision rule

Put dates only on pages where being wrong causes an incident or a compliance finding: runbooks, on-call escalation, data-residency statements, release and rollback procedures, anything an auditor reads. That is typically 3–8% of a corpus this size, 40 to 100 pages, which is one or two verifications a day and genuinely sustainable.

For everything else, instrument the reader instead of the author: an "this page was wrong" control on every page, with the report routed to the owning team and counted. A reader-reported error is evidence that the page is both wrong and read, which is the only combination worth spending time on. A page nobody reports and nobody reads needs deleting, not verifying.

It flips when the corpus is small. Under roughly 150 pages, a quarterly sweep by the owning team is cheap and the dates stay honest, because the work is visible enough that clicking without reading is socially obvious.

When this is the wrong answer

In a regulated context where an auditor requires evidence of periodic review, the date stamp is the control and you have to pay for it. The response then is to shrink the scope of what is in-scope for review, not to weaken the review: move the regulated statements into a small, separately governed set of documents, and keep the engineering wiki out of the audit boundary entirely.