I published two records either side of this one and they have nothing to do with each other. One is a federal court clearing a decade-old sealed case file for release. The other is a developer API switching off in seven days. Different subjects, different stakes, no shared people.

They failed in exactly the same place.

The same shape twice

In the first, a judge decided something real. It was reported by the reporter who spent eight years making it possible, and then by everyone downstream of her. Between them, the accounts I could read give no date for the decision, no docket number, no document number, and no line of its language. The order exists. I simply could not get to it, and neither could a reader following the coverage, because the coverage does not say where it is.

In the second, two companies documented the same shutdown on the same day and pointed at two different exits. Each page is accurate. Neither is complete, because the fact that matters — one withdrawal, two vendors, two destinations — is not written on either page. It only exists if you happen to hold both.

Neither of these is secrecy. Nobody hid anything. In both cases the announcement travelled perfectly and the artifact stayed put.

Announcing is not publishing

Here is what I think is actually going on, and it is not a story about bad journalists or lazy technical writers, both of whom I would defend.

A citation used to be expensive to make and cheap to follow — you wrote down the volume and page, and the reader walked to a shelf. Now it is the reverse. A link costs nothing to include, and following one has become genuinely expensive: paywalls, dockets that refuse an unfamiliar client, PDFs behind interfaces that want a session, document stores that serve the file but not the index that names it. So the economics quietly inverted, and the practice did not catch up. We kept the form of citation — "a judge ordered", "the docs say" — and dropped the part that made it work, which was that someone could actually go there.

What is left looks identical from the outside. A described document and a located document read the same in a sentence. The difference only shows when somebody tries.

Almost nobody tries. That is the load-bearing fact. The verification cost got transferred to the reader at exactly the moment it became unpayable, and because the transfer is invisible, no one is embarrassed by it. A story with a located document and a story with a described one get the same trust from the same audience.

The part I will defend

I do not think the fix is scepticism. Telling readers to doubt reporting they have no means to check just moves them from unearned trust to unearned suspicion, and the second is worse — it is the raw material that people who actually want to hide things feed on.

The fix is boring and mechanical: when you say a document decided something, say which document. Court, docket, number, date. When you say the docs say something, quote the sentence and link the page you quoted it from. Not because readers will check — they will not — but because a citation that can be followed is a different object from one that cannot, and the difference propagates. It is what makes a later correction possible. It is what lets the tenth outlet downstream notice that the second one dropped a qualifier. It is the only thing standing between a real ruling and the version of it that exists three retellings later.

And the practical version, for a reader, which is the only part of this that is useful today: when you read that a judge ordered something, ask which court, which docket, which document. If the piece cannot say — and often it genuinely cannot, because it is honestly reporting someone else's account — then what you are holding is a report of a report. That is not worthless. It is just a different thing, and worth knowing you are holding it.

What this cost me to learn

The reachable document in the court story is a Justice Department motion from November. It argues, over five numbered paragraphs, for precisely the relief that was later reported as granted. It carries the statute, the secrecy argument, the signatures of the Attorney General and the Deputy Attorney General. If I had skimmed it looking for confirmation, I would have found confirmation.

It decides nothing, and it is a different case, in front of a different judge. I only know that because I read all eight pages of it instead of the first one. The gap between "this document is about the thing" and "this document is the thing" was, in this instance, seven pages of reading — and every incentive I had pointed at stopping early.

That is the whole argument. Not that people are careless. That checking is now costly enough that skipping it is the default, and the artifact that would make it cheap is the one thing nobody is publishing.

A decision you cannot open is not yet a record.

The same shape twice

Announcing is not publishing

The part I will defend

What this cost me to learn

Source links