There is a particular kind of silence that follows a correction. The blog post is updated, the footnote appears, the changelog gets a line: corrected on November 3. The org feels honest. The record feels clean. And somewhere a customer is still making a decision based on the sentence that was deleted, because the correction was published to the same place the error was published to, and almost nobody returns to a page they already read.
A correction is not a new publication. It is an edit to a claim that people have already acted on. The error traveled; the correction has to travel the same distance, through the same channels, to the same people, or it is not a correction. It is a private note to yourself.
The test is not "did we fix the page"
Every team can fix the page. The page is the easy part — the version history is right there, the deploy takes a minute, the CMS has an edit button. What almost nobody does is measure whether the corrected claim reached the people who believed the wrong one.
The test is the same test as the disclosure layer, one level harder. A notice has to arrive before the recipient acts. A correction has to arrive after they acted, and then persuade them to undo it. That is a different problem: you are not informing someone, you are asking them to reverse a decision they made in good faith, and the burden of proof sits on you.
If a correction cannot name the sentence it kills, it is not a correction. It is a mood.
Name the sentence, quote it, kill it
"We have updated the post to clarify certain details" is the standard formulation, and it is nearly worthless. The reader cannot tell which claim changed, whether they relied on it, or whether the new text agrees with the old one. A useful correction does three things in the first two lines:
- Quotes the false sentence verbatim. Not a paraphrase — the actual string that was on the page. People remember the sentence, not the section.
- States what is true instead. One replacement claim, with the same specificity as the original. "The export retains records for 30 days" → "the export retains records for 7 days; the 30-day figure referred to the billing system."
- States who is affected and what they should do. If the error changed a decision, say which decision and what undoes it. If it changed nothing, say that too — "no action required" is a valid correction, and it is more credible than silence.
The version people quote is the version that gets cited, and it is usually the wrong one. Search results cache the old page for weeks. Screenshots outlive edits. A partner's deck contains the old number. Naming the sentence is how you give those copies a way to be corrected by the person holding them.
Reach is the same problem as disclosure, with worse odds
The disclosure layer already established that "sent" is not "delivered," and that the audience, not the sender, decides whether the message landed. Corrections inherit that problem and make it worse, because the audience for a correction is the subset of the original audience that read and believed version one — a group you can barely identify, let alone re-reach.
Three practical consequences:
- The channel that carried the error must carry the correction. If the number was in a newsletter, a blog footnote does not correct it. If it was in a sales deck, the deck has to be recalled, not annotated on a wiki.
- The correction has to be pushed, not posted. A new post at the same URL has roughly the reach of a new post, which is to say almost none among people who already read the old one. Push means email, in-product banner, direct message to the accounts that acted on it — with the specific sentence restated in the push.
- Timing has an edge. A correction issued before the decision is a warning; issued after, it is a request to reverse. Both are legitimate, but they are different messages and deserve different urgency. Sending the second as if it were the first is how corrections get ignored.
Two clocks, and only one of them matters
Write both down, every time:
- C1 — when the correction was issued. The timestamp on the edit, the send time on the email.
- C2 — when a reader who saw version one could act on version two. Not "opened the email." Could act: saw the correction, understood that it applied to them, and knew what to do differently. In their timezone, on the channel they used the first time.
The gap C1→C2 is the correction's real latency, and it is almost never in the changelog. A team that measures C1 will conclude it corrects errors in hours. A team that measures C2 discovers that most corrections land days later, if at all — or that they landed on people who never saw the original.
There is a third number worth recording when the correction concerns a decision: how many of the affected accounts actually reversed it. Password resets completed after a "your credentials may be compromised" correction. Orders re-quoted after a pricing error. Migrations re-run after a wrong cutoff date. That number is the only proof that the correction did work rather than merely exist.
The correction ledger
Six columns, one row per corrected claim — not one row per article. If a post contained two independent errors, it has two rows, because they have two audiences and two outcomes.
| Claim (quoted verbatim) | Where it lived (every copy) | When issued (C1) | Who saw v1 (and how we know) | When they could act (C2, channel) | What changed (action taken / reversed) |
|---|---|---|---|---|---|
The expensive column is the second one. The sentence lived on the blog, in the newsletter archive, in the PDF someone downloaded, in the sales deck, in the partner's integration guide, in the cached search result you do not control. One row per copy, exactly like the retention ledger — because the copy you forgot is the one still being read.
The column people skip is the fourth. "Who saw v1" is not knowable precisely, but it is estimable: newsletter opens, page views before the edit timestamp, the accounts that took the action the wrong claim recommended. An estimate with a stated method beats a blank.
The uncomfortable part
The incentive is to correct quietly. A visible correction costs trust twice — once for the error, once for admitting it — and a quiet edit costs nothing today. This is why most corrections are formatted to be findable only by someone who already knows what changed: a footnote, a date, a diff link.
But the quiet correction is a bet that nobody will notice the gap, and the bet loses on a long enough timeline. The customer who acted on the wrong number finds out when it hurts, and at that moment the correction they never saw becomes evidence that you knew and did not tell them. The footnote that felt responsible becomes the exhibit.
The honest formulation is uncomfortable and short: this sentence was wrong; here is what is true; here is what it cost you if you acted on it; here is what we are doing about that. Anything less is a record-management exercise.
Friday version
Take one correction you have issued in the last year. Quote the wrong sentence verbatim. List every place it lived — including the PDF, the deck, the cached page. Write C1 and C2 as two different moments; if you cannot name C2, that is the finding. Then answer one question: did anyone who acted on the wrong version find out? If the answer is "we assume so," the correction did not land. Send the push this week, to the actual list, with the sentence restated.
What comes next
The correction layer closes the loop on the disclosure layer: first you prove the notice arrived; then you prove the fix arrived to the people who read the broken version. The next layer is the one underneath both — the retraction: not "we corrected a sentence" but "we withdrew a claim and everything built on it." A correction edits; a retraction has to unwind decisions, contracts, and downstream documents that copied the claim. That is where the audit trail either holds or snaps.
If you are holding a correction you issued but cannot prove reached the people who read the wrong version — a "we have updated the post" line, a fixed footnote nobody re-read, a number you changed after three customers had already quoted it — say so in the comments. We keep a one-page correction ledger (the wrong sentence verbatim, every copy it lived in, when it was corrected, who saw v1, and who actually found out) for exactly that, and I will send it over. No form, no email list.
Top comments (0)