Corrections

Everything below was published wrong and then fixed. Each entry says what we had, what is actually true, and how it surfaced. Wording changes and typos are not logged; anything that would change what a reader does is.

5 September 2026 — the headline count was one short

We said: that twenty-six of fifty-one regulators publish guidance contradicting their own rules — in the comparison table’s heading and lead, in the state checker, and on two other pages.

Actually: twenty-seven. The same page enumerated all twenty-seven immediately below the heading that said twenty-six. The figure had been typed by hand and incremented as states were added, and one increment was missed; the list stayed right while the number drifted.

Found by: counting the list against the heading before reusing the figure elsewhere. The number is now derived from the list rather than typed, so the two cannot disagree again — and each of the twenty-seven was re-checked against its own page before the figure changed.

5 September 2026 — dark-mode previewer gave backwards advice

We said: the tool flagged “no explicit text colour” as worth changing, and told readers to set an explicit colour on every text element.

Actually: that is what causes the failure the tool exists to catch. An explicit colour is one the client honours — so color:#1a1a1a is precisely what produces dark text on a dark background. Unstyled text inherits the client’s own colour and usually adapts, which is often safer. The tool now parses every explicit colour, computes its luminance, and flags a locked-in dark one as the real problem.

Found by: writing test HTML to hand to a reader. This is the only entry here where the code worked correctly and the advice was wrong, which is the worse kind to ship.

5 September 2026 — HTML checker missed a linked SVG

We said: nothing — that is the problem. The SVG check matched an inline <svg> element only.

Actually: a logo referenced as src="logo.svg" is the far more common case, since an SVG is what a designer hands you, and Gmail strips it just the same. It passed silently, which reads as a pass rather than a gap.

Found by: writing a realistic broken signature to test with, rather than a synthetic one.

5 September 2026 — Alaska’s guidance was moved, not withdrawn

We said: the Commission’s “Best Practice – Advertising” document had been taken down, and Alaskan licensees had no official advertising guidance to consult.

Actually: it is live. It was relocated into an archive subfolder with no redirect, so its published address returns a 404 and search engines still index the dead one. We had fetched the published address, got a 404, and inferred removal. A different document really is missing — a 2022 Teams paper, and it is the only Alaskan text that ever named email signatures. We also had one divergence backwards: the guidance is more lenient than the rule on home-office addresses, not stricter.

Found by: a verification pass that located the relocated address. We fetched it ourselves before correcting.

4 September 2026 — “every state requires the brokerage name”

We said: on the comparison table, for forty-eight jurisdictions: “the one thing every state agrees on: your brokerage or broker’s name must appear.”

Actually: Montana does not require it. Its rule asks for your own name plus a statement that the advertisement is made by the licensee or a brokerage company — the brokerage need not be named. The claim was rewritten to lead with the exception rather than quietly dropped.

Found by: researching Montana, which was the forty-ninth jurisdiction added.

4 September 2026 — the state checker repeated that claim to Montana users

We said: the interactive checker printed “Your brokerage or broker’s name. Required in every state checked so far” for every jurisdiction, Montana included.

Actually: false for Montana. The line was hardcoded rather than derived from the data, so correcting the table did not correct the tool. The exception is now held as data, and the build fails if the exception set is ever anything other than what the pages say.

Found by: a live functional check of the tool after the series completed. No validator could see it — the markup was perfectly valid.

4 September 2026 — Delaware’s phone-labelling rule was not unique

We said: Delaware’s requirement to label every phone number by type was “a requirement we have not seen in any other state”.

Actually: New Jersey carried a near-identical rule — the same sentence stem, the same “Res.” example, plainly cognate text — and repealed it on 20 January 2026. Delaware kept a rule its neighbour had just dropped, which is a different and more interesting claim than the one we made.

Found by: a verification pass against the full text of Pennsylvania, Maryland and New Jersey. Three of the four claims held; this one did not.

4 September 2026 — Oklahoma’s 2026 amendment stated too firmly

We said: that an amendment had been adopted with an effective date of 1 November 2026.

Actually: final adoption could not be confirmed from the published record. The page now presents the date as expected rather than settled.

Found by: a second check that could not corroborate the first. Overstating a future effective date is the exact failure this series documents in other people’s guides.

4 September 2026 — Kentucky’s team-name rule had been rescinded

We said: that a Commission resolution barring “realty” and “real estate” in team names was current working policy.

Actually: it lasted 28 days. Passed 21 May 2020, rescinded 18 June 2020, never reinstated. The announcing memorandum returns a 404 because it was withdrawn — and we had read that dead link as a rule that was merely hard to find. Kentucky also moved category as a result, taking the count of regulators contradicting their own rules from eighteen to seventeen at the time.

Found by: a check of Commission meeting minutes made after the page was already live. It is the reason every research brief now says to read minutes, not just rules — and the reason we treat a 404 as evidence of nothing until the document is located.

What we have learned from these

Two patterns, both now built into the method:

• A dead link proves nothing. Kentucky’s 404 meant a rule really had been withdrawn. Alaska’s meant a folder had moved. Both were read wrongly the first time, in opposite directions. We now locate the document before drawing a conclusion from its absence.
• Validation is not verification. Four of these were in perfectly valid markup that passed every automated check. They were found by using the thing — picking a state, pasting a signature, dropping in an image. Every tool now gets a functional pass before it ships.

If you find something wrong on any page here, tell us at hello@autographs.app and it will be checked, fixed and added to this list.