The first thing we look for on a page is the date it was published. Not the headline, not the author, not the first paragraph. The date. We scroll until we find it, and if we can’t find it, we already know something important about the source.
This habit takes a few seconds. It rearranges everything we do with the rest of the page.
What the date actually tells us
The obvious thing a date tells us is whether the source is current. That part is true and matters. A page about API rate limits from three years ago will describe a system that has probably changed, and reading it as if it described today’s system is one of the easier ways to get a wrong fact into a finding.
The date does more than mark currency, though. It tells us which era the claim belongs to. A statement about how teams handle deployments in 2020 is not the same as a statement about how they handle them now, even if the words are identical. The same paragraph, written four years apart, sits inside very different assumptions about what tools exist, what costs what, and what counts as standard.
When we read a source out of that context, we tend to absorb its assumptions without noticing. A “best practice” from 2019 reads as a best practice today, because the article doesn’t usually say “this advice is conditional on the way our toolchain looked when we wrote it.” It just says best practice. The date is what we have to put that conditioning back in.
What we do with an old source
We don’t throw out old sources. Most of what is useful is not bleeding-edge. A piece about how a protocol was designed, why a system has the shape it has, what tradeoffs were considered ten years ago, can be exactly the right thing to read even when it’s old. Sometimes especially when it’s old.
What changes is how we treat what we read. From a current source, we might pull a specific number or recommendation and place it directly into a finding. From an old source, we are more likely to pull the underlying reasoning and look for a current source to confirm whether the specifics still hold. The old article tells us what to look for. The new one tells us where we are.
This sounds tidy on paper. In practice it requires a kind of split-screen reading: holding the article’s claims in one column and the question “but is this still true?” in another. We have learned that the date is what reminds us to keep that second column open.
The trickier cases
Most pages either have a date or they don’t. The honestly dated articles tell us the date. The honestly undated ones don’t pretend.
Then there is a third category, the one we have learned to be most careful with: pages whose dates are misleading. There are a few common shapes.
Updated pages. A documentation page might say “Last updated: last month” while the substantive content was written years ago and never revisited. The freshness signal is real for the page metadata and false for the content. When we see a recent “last updated” we still look for changelog notes, inline annotations, or “as of vX.Y” qualifiers. A recent update timestamp on an old article often means someone fixed a typo, not that the underlying claim has been re-verified.
Evergreen pages. Some publishers strip dates from their content as a deliberate choice. The page looks current because nothing tells us otherwise. The instinct here is to assume freshness, which is almost always wrong. When a page is undated we treat it as “of unknown era” and look for clues: tool versions mentioned, prices quoted, links that lead to retired services. The clues usually place the page within a year or two.
Archive captures. Sometimes we find what we want on a web archive, and the only date we have is the date the archive captured it, not the date the article was originally written. The capture date tells us the latest the article could be from. It tells us nothing about how recently it was written. We try to find the original timestamp inside the captured page and treat the capture date as a ceiling rather than a stamp.
Republished or aggregated pages. Aggregators sometimes put a current date on something that’s just a syndication of an older piece. The byline and headline are unchanged. The publication date is set to when the aggregator picked it up. The visible date can be days old when the actual content is from years prior. The byline date, if there is one, is more reliable than the page’s surface date.
Why the date is the first check, not the last
It would be possible to read the source first and check the date later. We have tried it. It doesn’t work as well, because by the time we have read a well-written article we are already partly persuaded by its claims, and discovering at the end that it’s seven years old produces a strange kind of cognitive backtracking that we don’t always fully unwind.
Reading the date first sets the frame. If the article is from this month, we read it like a current report. If it’s from five years ago, we read it like a historical document. The same paragraphs land differently inside those two frames, and the difference is the whole point.
The other reason to check first is the cost of not. The cost of reading a stale source as if it were current is sometimes small, a sentence we delete from a draft, and sometimes large, a recommendation that no longer applies. We have shipped both kinds of mistakes. The ones where the date was actually visible on the page, and we just didn’t look for it, are the ones that bother us the most.
What this changes about the rest
The date stamp turns into something like punctuation. It tells us how to read the sentence that follows. Without it, we can still read the sentence, but we have to do extra work to figure out what era it’s from, and we don’t always do that work consistently.
The version of this we have arrived at, after many findings that needed correction, is that we now feel slightly uneasy reading any page where we can’t find a date. The unease is useful. It keeps us from treating undated assertions as established fact. We can still use those pages, but we annotate the finding accordingly: “source undated, content appears to refer to roughly the X era.” That sentence sounds weaker than the alternative. It also happens to be accurate, which is the only standard we are trying to meet.