A plain account of how this site has changed and why, newest first. It is kept so that anyone returning after a few weeks — including the person who writes it — can pick the thread back up without reading a commit history. Each entry says what changed and what it was for; the machinery is in column C where it belongs.
This site is the technical reference for the GYRO format, and it is written in the format it describes: plain answers on the left, every source previewed in the middle, technical detail on the right.
It currently carries twenty-three previewed sources, a search box that indexes the page from its own contents, and three sub-pages — D, the setup sheet a subject never has to see; E, standing up a domain; and F, what owning one of these sites actually consists of.
Two sample sites live under it, at /sites/: an invented farm and an invented tree service, both marked SAMPLE, both built exactly as a real one would be.
Every page of every site now has a copy elsewhere. All 32 pages across the four live sites are sent to the Internet Archive by a command on the author’s machine, dated, and each capture is verified to exist rather than assumed from the reply. What that is worth, and the three things it is not, are answered in column A.
Every page now carries a threshold and a set of switches. A double rule with the 3Col mark knocking a gap in it separates what belongs to the subject from what is the same on every 3Col site; below it, the reader decides which of the three columns are on the page at all.
The argument has not changed since the first day. One topic, one HTML file, no server, no database, no subscription, no vendor — and every claim on the page backed by a source the reader can check without leaving the column.
Walter said the gold looked a little too dark. Measuring it before changing it turned a matter of taste into a matter of fact: as text on the paper background that gold sat at 2.50:1, against 4.79 for the green and 5.71 for the red. It was already failing, and lightening it to taste would have made it worse.
The cause was one value doing two unrelated jobs — marking column B in bars, borders and icons, where it wants to be bright, and setting the letter B and the tab key, where it wants to be dark. Those were separated. The mark went brighter, the lettering went darker, and both now pass. Every icon file followed, recolored by solving each blended pixel against the paper rather than by a flat swap, and checked by census: no pixel of the old value survives anywhere in the family.
Worth stating plainly, because it explains a design decision that otherwise looks arbitrary. Printed on a monochrome laser the green and the red land at 103 and 91 — twelve levels apart out of 255. On paper they are the same color. That is why column A is always leftmost and column C always rightmost: position carries the code, and hue only confirms it.
And a citation that led nowhere. A reader who had switched the middle column off, then followed a source numeral from A or C, arrived at nothing. The machinery to reopen a switched‑off column was present and correct — but wired to an event the page never fires, because the numerals are not links. Following one never changes the address, so the browser never announces anything. Fixed across every page that has columns. Then the test was tested: with the fix disabled the check reports the reference as hidden and zero by zero, which is exactly the fault described. A test that cannot fail proves nothing.
Walter: “in all instances it should match below between the two horizontal rules.” Checked by measuring the drawing rather than by looking at it, which found three faults where one had been reported: a favicon rotated a quarter turn, a home‑screen icon rendered from that same rotated file and inheriting the error, and a third set still carrying an older shape in the wrong colors entirely.
What was already right was confirmed rather than assumed. A comparison pulled the geometry out of every page’s threshold mark and out of that page’s favicon and compared them field by field — identical across all six, the same drawing by construction rather than by coincidence, since the mark takes its colors from the page’s own variables.
A gap turned up while looking that nobody had asked about. Only two pages in the whole family declared a home‑screen icon; fifteen had none, so saving one to an iPhone home screen would have produced a screenshot of the page instead of a mark. They all have one now.
Walter wanted the column switches on every 3Col site. Surveying first was worth the delay: only seven pages in the family carry three columns at all. worland‑dog‑law.online and baconian‑ciphers.online do not lack the switches — they have no three‑column pages for switches to control. Their pages are an audit, a chronicle, an ordinance text and a research record.
The six that remained posed a real design problem, because they already had an A/B/C tab strip. Two controls doing similar things would have been the wrong answer. Instead they were made orthogonal, and the distinction is worth stating because it is now part of the format: the switches decide whether a column is on the page at all; the tabs go on deciding which one has focus. A column switched off leaves the grid and the tab strip together, and on a wide screen the grid closes up around what is left. Turn all three off and the page is its masthead and the band — which is still there to turn them back on.
Almost all of it is ordinary CSS responding to the state of a checkbox, so it works with scripting off. Live at 00:35 on the 27th, and verified by fetching each published page rather than by assuming the upload had worked.
Walter: “above the line it is freeform to whatever the app requires, below it, the 3Col per se.” Built as a contract rather than a divider. A double rule — nothing else anywhere in the family is doubled, so the doubling says different in kind before a word is read — with the 3Col mark knocking a gap in the line, and beneath it a caption: Everything below is the same on every 3Col site.
The mark itself was then convicted by the person who commissioned it. Walter, looking at a generated version: “your three bar icon looks to me like the flag of Canada with a maple leaf in the middle.” He was right, and it condemned the one already shipped that afternoon. Three tall bands with a device in the center is a flag, not a diagram. The answer was not fewer bars but something above them — a masthead block, then the rule, then the three columns — so that the mark draws the format’s own thesis, and the threshold on a real page is the same figure assembled from the page itself.
The columns also got a code, and Walter supplied the semantics unprompted: “green for public outreach, red for prior knowhow, yellow intermediate, like a traffic signal.” The metaphor carries its own warning. A real signal never trusts color alone — red is always at the top and green always at the bottom, because red and green are precisely the pair that fails. Tested against simulated red‑green color blindness, the first palette’s green and red collapsed to nearly one color. The tuned version keeps every hue exactly as drawn and moves only lightness, and position does the real work.
Walter: “isn’t some brief copyright notice needed at footer?” Strictly, no — copyright has been automatic since the United States joined the Berne Convention in 1989, and a notice adds no rights. But it names the owner and the year, and under US law a notice on published copies takes away an infringer’s plea of innocent infringement in mitigation of damages. So it earns its line.
The question could not be answered on its own, though, because of what was sitting in the repositories. Two of the four carried a bare MIT license at the root, which granted the whole world the right to copy and reuse the prose as well as the code — the defect written up on Page F two days earlier and deliberately left standing. A footer reading © Walter Wilkinson above a file granting everything away is worse than no notice at all: two opposite claims in the same repository, and a reader entitled to rely on either.
So the notice and the defect were one decision, and Walter took the obvious one: split them. The license file now names the two things before it grants anything. The format, markup, stylesheets, scripts and tooling stay under MIT, unchanged — the method has to stay free or the whole argument for it collapses. The written content is reserved: the prose, the answers, the source previews and the reasons they are cited, the chronicles, the maps and the illustrations. Quotation with attribution is fair use and is welcomed in as many words.
Two of the four repositories had no license file at all. That is the stricter position by default — nothing granted — but it says nothing out loud, and a subject handed such a site has no way of knowing where they stand. They now carry the same split as the rest.
Every page in the family, 34 of them, now ends with the same line: © 2026 Walter Wilkinson · code MIT · words reserved. On the sites that can reach it, the last three words link to Page F.
Page F itself was rewritten rather than tidied. The defect stays on the page, with what it was, when it was found, when it was corrected, and the two-day interval named explicitly — because anyone who copied a sample page’s text in those two days did so under a grant that was really given, and nothing here pretends to take it back.
It came out of a question Walter asked about something else entirely: does the link checker handle internal links, or only external ones? Answering it turned up that internal links are checked by nothing at all — and then a second question followed from it. What about ibid., the footnote convention: is it still needed here?
It is not, and the reason is worth more than the answer. Every reference on these pages names what it points at, never where it sits. A numeral means the same thing however the reader arrived — from the top, from a numeral in the third column, from a search, from somebody else’s link into the middle of the page. Ibid. means “the note immediately before”, and on a page read that way there is no before.
Walter’s own summing up was “yet another advantage to this, logical simplicity”, and that is right, with one refinement now written into column A: the simplicity is the consequence, not the aim. What is actually being refused is relative addressing. Four things follow from that refusal, and until today all four were being enjoyed without anybody naming the cause:
The Chicago Manual of Style reached the same conclusion for print at its seventeenth edition, discouraging ibid. on the ground that it “risks confusing the reader”. Its citation chapters had to add permalinks and electronic page numbers in the same revision, which is a fair indication of what forced the change. That is source 23.
This is recorded as a principle rather than a preference because it is the thing to protect first if the format is ever altered. A change that makes any reference depend on position would cost all four benefits at once, and would do it silently.
The format’s whole proposition is do not take my word for it, check. Until today that sentence had one hole in it: everything a reader could check was served from an address the author pays for, on a host whose terms the author has to accept, and nothing outside that arrangement recorded what any page had ever said. Walter asked whether there was an off-the-shelf way to fix it. There is.
A new command, ARCHIVE SITES, finds every site on the volume the way the link checker does — a folder with both a CNAME and an index.html — turns each page into the address it is served at, and submits it to the Internet Archive’s Wayback Machine. Thirty-two pages across four sites, and new sites are picked up without anyone editing a list.
Three decisions are worth recording because each one could have gone the other way:
The answer in column A sets out what this buys: a date the author did not set, proof against quiet editing, a citation that still resolves when the domain does not, and — with a free account — the cited sources captured alongside the page that cites them. It is also honest about what it is not. Not a backup: the archive keeps the page as a reader sees it, not the files it was built from. Nothing for search ranking. No visitors.
The reason it earns a place in the format rather than being a convenience: an archived copy is the one link in the chain the author does not control.
Walter, reading all of this back: “I don’t remember if Transparency is one of the many advantages offered by this project, but surely that belongs somewhere.” It was not. The word appeared nowhere on any site in the family — the only match anywhere was a line of stylesheet setting a shape to transparent.
That is worse than an omission, because transparency is not one advantage among several here. It is the property every other rule is an expression of: sources previewed in place, a failure recorded in every column C, a defect published rather than fixed quietly, an argument dropped for being uncheckable, checks named rather than merely run, dead links marked rather than deleted, samples that say they are samples, and this page. Eight rules, one property, never named.
It is named now, in column A, and framed the way it has to be framed to mean anything: every one of those rules costs the author something. That cost is the whole of their value.
Until today the record of what had changed here was two one-line cards at the top of the middle column, and a private working log that is addressed to a machine, runs to 71,000 characters, and is wrong about its own status the moment anything is published. Neither is a record a person can return to.
This page is the third internal link, and it sits last in that block: the two most recent changes first, then the whole story. It was very nearly called a Synopsis; Chronicle is the better word, because a synopsis is a condensation and what this is, is a dated record. The block at the top is the synopsis; everything under it is the chronicle.
Every page in this family had been saying a subject owns their site outright. That sentence appeared once, with nothing behind it. Page F now takes it apart: you own the domain and the words, you do not own the hosting, and the reason that is survivable is that everything you own is portable in an afternoon.
Writing it turned up a real defect. Every repository here carries an MIT license at its root — right for the format, which is meant to be free — but the file sits above sites/ as well, so the grant reaches a subject's own words and prices. Harmless for two invented samples; a genuine mistake for a real business at handover. It is recorded on Page F as a live defect and has not been fixed, because the format asks that failures be published rather than tidied away.
“Why not just keep my Facebook page?” is the first thing anyone says about all of this, and until today the reference did not mention Facebook once. The answer now rests on three things a reader can check rather than on the usual folklore about link reach, which is unsourceable and is explicitly declined on the page: a feed is ordered by recency so an answer sinks whether or not it is still true; automated clients are asked not to fetch the pages at all, so nothing can archive, verify or cite them; and the address belongs to somebody else. The recommendation is both, never migration.
The farm sample showed the format carrying a place. The new tree service sample shows it carrying a problem: a stranger standing in their own yard cannot tell whether a contractor is insured, and the usual proof is a folded certificate in a glovebox. So its first answer is the insurance, and every line of it links to a check somebody else keeps.
Separately and more embarrassingly: both samples have lived at this domain for days while nothing here linked to them. Only the grant proposal did. Column A now lists them.
Column A's answers are folded shut, and a browser's own find bar does not reliably look inside a folded answer — so a reader could be told a word was absent when the page plainly contained it. A search box now builds its index from the page's own contents at load, opens the right column, unfolds the answer and marks it.
The same day, the Internal links block went from one recent change to two — one was too thin to be a record — and the passage about sources arriving at the reader was split in two, so that the Back button carries equal weight with the card that comes to you. Both ends of the interruption, not one.
The strongest claim this format makes is about cost, and it turns on communication rather than typing: one person means zero lines of communication, and the cost of more climbs like a hockey stick. The page now argues it with a drawing — the connectors across the vertices of regular polygons — and pairs it with a list of the industry job titles the arrangement simply removes.
Also: “publish” had been doing two jobs, meaning both stand a site up and send a change live, which made the GitHub answer needlessly frightening. Split apart, with new answers on changing the words later and on who sets a site up in the first place.
For a while this page described less than the sites built from it actually did, which is the wrong way round for a reference. The format's own features were surveyed from the live sites and written in: the quality-assurance checks, named on the page rather than merely run; Phone view shown to every reader here, because on this site the layout is the subject; and the sites/ folder, so samples live at this domain rather than costing a new one each.
Phone view built as a feature of the format and fitted to every site at once; a whole column made clickable as its own button; reference numerals set on white discs so they read as controls. A share card found to be serving the wrong site's image and restored, then replaced by one generated from three fields instead of drawn by hand — the first thing here to stop being a manual job.