This week ★ stuff added eight zines, a field guide, three Foundations pieces, a listening rack, two editions of a new weekly digest, and three new collections — and what it kept finding was a condition that had been printed right there in the source and fell off somewhere on the way to daylight. Entropy tends to a maximum in an isolated system. Solar is the cheapest electricity in history for projects with low cost financing that tap high quality resources. A dandelion root opens a channel seventy micrometres wide. The UN’s own resolution says plainly that no single projection can perfectly represent the curved surface of the Earth on a plane — and every outlet reported the vote as a swap to a more accurate map. On this site the week went to governance: our bylaws published for the first time and published before the board votes on them, a conflict of interest form made public over a recommendation to hide it, what it costs to sit on our board written down including the parts most organizations leave you to discover after you say yes, and a transparency log carrying the missed board meeting and the vacant Treasurer seat alongside the wins. Plus two new Why Sheets — Monotropism at Fergus Murray and Helen Edgar’s request, and Masking and Burnout, written for a family whose son was refused an evaluation because he is doing fine. A condition that travels attached to the claim is a different object from a condition somebody has to remember. So is a check that measures the thing you actually care about.
Throughline
- The qualifier is not lost in the literature. It is printed on the equation, and it comes off on the way out. Rainbow Serpents and the Laws of Thermodynamics opened its own list of unfinished business by admitting its thermodynamic definitions came from teaching sources rather than a named authority — and the primary that closed it handed the essay its thesis in miniature. Singh & O’Neill’s 2022 review in Reviews of Modern Physics prints the entropy inequality with the parenthetical label (isolated system) attached to it, then introduces the non-isolated form by noting that the climate system exchanges energy with space “and it is therefore not isolated.” The condition is not in a preamble or a footnote. It is a label on the formula. Then the same shape, five more times, in five different fields. The IEA wrote that “for projects with low cost financing that tap high quality resources, solar PV is now the cheapest source of electricity in history”; what travels drops fourteen words about capital cost and site quality — most of what sets the price for a technology that is nearly all up-front cost — and quietly turns electricity into energy, extending a claim about 22% of final consumption to all of it (None of Those Miles). The dandelion-in-compacted-soil claim kept the verb and dropped every condition: 6.9% porosity in the first 70 µm, 3.9% by 140, and nothing at whole-column scale (The Width of Itself). The soil-underpants headline is 59% ± 15%, printed flat twice in our own draft, with its own authors saying treat with caution (Came Back Clean). A review chapter reports the Arizona agrivoltaics experiment as chiltepin and tomato up 150% and 90% and drops the jalapeño entirely, which is the one result that complicates it (Two Peppers). And Which Distortion found the UN’s operative paragraph claiming only “a more accurate representation of the sizes,” from “a relevant option,” “where their sizes are critical” — three conditions in one clause, none of which survived contact with a headline. In every one of these the source was more careful than the coverage. The failure is transmission, not scholarship.
- A check that measures the wrong property passes, and hands you a number while it does it. The Width of Itself turns on a measurement in micrometres, and the cells rendered as “0–70 MM” — the micro sign sat inside a label carrying
text-transform: uppercase, uppercasing µ yields Greek capital Mu, and in Space Mono that is indistinguishable from a Latin M. A factor of a thousand, and all eight gates passed it twice — none of them wrong to, because the markup is well formed, the class is live, nothing collides, and a wrong character has exactly the same contrast as a right one. The week is full of this. A new card shipped inside another card on the front page, which is valid HTML with balanced tags — and cost 151 search records where there should have been 152, so the new card’s text was not independently findable. Every page got a<main>landmark in the morning and by the afternoon twenty-four of them had a landmark pointing at nothing: the front page’s opened inside the masthead and closed one line later, measured live at 0 characters, 0 children, 0 of 137 cards — and the new “exactly one<main>” check passed all twenty-four, because exactly one is true of a landmark that contains nothing. Glimmer Wire‘s first edition printed “7 records 100%” — a coverage figure computed on a code path the builder had already skipped, leaving the standing notice, every held item and every refusal unfindable. A later page reported 3000% coverage on 34 records. The field guides mirrored 8,223 characters of real argument and eleven of thirteen entry names absent — a page that looks mirrored and is missing its substance, which is worse than one that visibly fails, because nothing ever complains. And oEmbed returned HTTP 200 with the right title and channel for videos that will not play: it confirms identity, not existence.* The fixes were not more checks. They were checks that ask the property that matters* — is the page’s own<h1>inside the landmark; is a card-wrap opening inside a card-wrap, counted by depth rather than matched by string; is the indexed text of a page four to nine times its static text, which separates client-rendered guides from every server-rendered page with no overlap at all**. - The corrections that cost the most this week were ours, and three of them killed the piece that found them. We published that a
<br>with no space beside it makes a heading announce as one run-together word across 163 headings on 48 pages — and it does not: Chrome inserts the whitespace itself when computing the accessible name, in every shape we could construct. The 347-edit sweep was written, measured, and discarded rather than shipped. We filed a critic’s charge that a fibromyalgia paper had been overreached by headlines, and on reading both papers in full found that sentence is the paper’s own abstract — arguing with a paper and arguing with a press release are different jobs, and we filed the wrong one — so the piece that was going to come out of it is parked, with exactly what would unpark it recorded. We described the ordinary Hall effect of 1879 while reporting a paper about the anomalous Hall effect, reproducing the press release’s conflation in the act of complaining about it.* We credited an elephant with a result from a different species — every figure on the card is the African elephant and the mirror study is three Asian* elephants — which is not pedantry, because male African elephants often touch their own heads owing to temporal gland secretion, and the female Asian elephant’s vestigial gland is precisely what let the authors rule that out. Carrying a finding across species reintroduces the confound the species choice removed.** A card claimed corvid self-recognition on five hand-raised magpies, two significant, with a failed replication since. A Kanzi card written from a search summary was missing the distinction the authors were careful about: comprehension, not production. And Which Distortion opened its hero by asserting the General Assembly “voted 164 to 1 to correct the world map” — four screens above a sentence calling this a choice rather than a correction. Ryan caught that one, as he caught four errors in Rainbow Serpents by reading the shipped page, and a card nested inside another card after the push. - Publishing the governing document before the vote costs nothing, and a log that only records the flattering things is marketing. Our bylaws went up for the first time — all forty-two sections of the document executed in 2022 — with one sentence deleted from §2.1: the clause requiring every director to be a member of the founder’s family, by blood or marriage. Marked Proposed, not adopted, because the board does not vote until this month, with the removed sentence quoted in full so nobody has to diff two documents. Publishing early is not generosity: §6.1 only permits a bylaw amendment where notice went out with the notice of the meeting, so it had to be visible anyway. Letting everyone else see it at the same time cost nothing. The same week we published a compliance form on an open URL over our own recommendation to keep it private — that reasoning conflated a public page with public answers, and Ryan said so — reproduced an article-numbering error in the executed bylaws rather than tidying it, corrected sitewide that our President had been published as “Board Chair” because we had written the routing of two policies from our own website, and logged publicly that we built a retention policy from a consultant’s sample and only afterwards found counsel had written us one in 2022, sitting unread in the formation binder. Writing staff and volunteer onboarding then surfaced a gap nobody had gone looking for: Article V indemnifies directors and officers, not volunteers, so our unpaid Co-Creative Directors may have no protection at all. That is now an eleventh board decision, and it is on the log.
The Work
The Condition Was Printed on the Equation
Rainbow Serpents and the Laws of Thermodynamics is an essay in Foundations, and the physics is exact. An isolated system exchanges neither matter nor energy. A closed system exchanges energy but not matter, and its entropy can decrease. An open system exchanges both, where internal entropy production stays non-negative while total entropy can fall, because the system exports it. That is not an exception to the law. It is how anything alive works. Tyson Yunkaporta (Apalech clan, western Cape York) argues that applying the second law to open interconnected systems is a curse — in the precise sense he defines, “a true law or pattern applied where it doesn’t belong” — and the essay checks the physics half and finds it correct.
The traffic runs one way and the page says so. This is a criticism of our physics from outside our discipline, which we checked. It does not claim Aboriginal Law is validated by thermodynamics, and it does not explain the Rainbow Serpent — not one being, not one nation’s, not ours — quoted from Yunkaporta and left there, with the commodification history named as the reason for the restraint. Three infinities are held apart with lineage explicitly refused: the ouroboros, the Rainbow Serpent, and the autism infinity loop. They rhyme. Rhyming is not descent. And the first correction is to ourselves: No. 43 says the second law governs closed systems when the precise word is isolated, corrected in the same breath as the clarification, because we are not entitled to do the second without the first.
Then Ryan read the shipped page and found four more, every one of them a fault the essay is itself about. A manufactured number, in the first sentence — the missing qualifier was called “six words long” and never named; it is four, of an isolated system, and the count was a flourish written without counting, which is the exact fault of No. 79, committed in the opening line of a piece about dropped qualifiers. The opening now prints the law with the qualifier and without it, so the reader sees the loss rather than being told about it. The essay never answered its own question — it argued the running-down picture is misapplied and never said what happens instead, so a new section says it: Earth absorbs energy carried by low-entropy photons from a ~6,000 K source and radiates it as high-entropy infrared at ~250 K, energy in and out nearly equal, far more entropy leaving than arriving, and every ordered thing paid for out of that difference. Mechanism given, number withheld, because the commonly quoted export ratio depends on which fluxes are compared and published treatments differ by a factor of two or more. A correction that ran the wrong way — we had said Yunkaporta writes closed where the precise word is isolated, implying a shared imprecision. He does not share it: describing what the doctrine demands, he writes “systems must be isolated and exist in a vacuum of individual creation,” the stronger word, in the right place. The imprecision was ours alone. And the closing line denied that the universe’s total entropy increases. It does.
Closing the open item on Singh & O’Neill vindicated the one decision the essay had been least sure of: the review says “vastly more” and prints no multiplier either. The editorial pass that followed retitled the section carrying Kaya Oldaker’s The Autism Infinity Dragon* — the first artwork on the site by another artist, public domain by the artist’s own release and credited anyway.* It had been “The dragon that is not a serpent,” sitting directly above the image with a second negation as its first sentence: three refusals before anything positive was said about the work.** It is now “The head does not take the tail” — the actual finding, verified by cropping the original around the head before a word was written about it — which sends the reader into the image rather than against it, with the conflation note moved to the end and reframed as ours to manage. A settler-run essay has no business annotating an Autistic artist’s work with a disclaimer about somebody else’s cosmology. And fixing that exposed something worse two inches below: the next heading read “a closed-system model of a person” while its own body said isolated — the exact conflation this essay corrects No. 43 for, committed in a heading on this page. Four loose usages found, all fixed, six remaining instances each checked and each correct. One process note stayed: a scripted fix silently failed to save two of the four, and only the after-the-fact audit caught it. Audit the file, not the script’s success message.
The same shape then arrived in four more materials. None of Those Miles derives the solar learning rate rather than quoting it — 50 observations from 1975 to 2024, a 497-fold fall in price against 21.7 doublings of capacity, giving 23.5% off per doubling by regression, 24.9% endpoint-only, about 20% as usually stated — and prints all three while calling none of them the number. Iterating the series then contradicts the word law: prices rose in 1989, 1990, 2005, 2006, 2007 and 2022, the last by 10.2%. Six reversals inside a 497-fold decline is a trend with exceptions. Two of our own figures were wrong in the same paragraph that argues qualifiers matter — a strait given as “twenty-one miles” is 21 nautical miles, and electricity called “roughly a fifth” is 22%, rounded the wrong way.
The Width of Itself (No. 83) is built on Burr-Hersey, Ritz, Bengough and Mooney in J Exp Bot, and its title is its finding: the dandelion does not fix the lawn; it opens a channel the width of itself. Prompted by Nick Covington’s “Every Dandelion is a Triumph” for Human Restoration Project, it sharpens his argument rather than disputing it — triumph is the right word, solution is not, and most dandelions are apomictic clones, so the grass-versus-dandelion reading has a clone-field on both sides. The closing spread turns on a second dandelion: in Boyce and Ellis’s borrowing from Swedish the dandelion child is the low-sensitivity one, and the orchid half is not fragility, because in good conditions orchids do better than dandelions — so every orchid failing in a compacted room is also a result nobody got to see. Both dandelions agree about the ground: hard ground does not produce variety, it selects for whoever needed least from it, and the survivors get counted as the normal range. Three source corrections went with it, one of them a shape this site has shipped before — the good-neighbours lawn line is Abraham Levitt’s, via Steinberg, not Elizabeth Kolbert’s; she is quoting it.
And Which Distortion was written the day after the General Assembly adopted A/80/L.104 by 164 to 1, with 6 abstentions. Three equal-area projections were computed from their published forward equations rather than drawn: Gall-Peters, Eckert IV and Equal Earth each enclose exactly 4π square units with det(J) = 1.000000 at all 35 sampled points — and each pays a different price in shape. Gall-Peters, the map most people mean by “the fair map,” distorts 2:1 at the equator, the latitude the campaign is about, which is why three cartographers built Equal Earth in 2018 after Boston Public Schools adopted Peters in 2017. France announced Eckert IV, not Equal Earth, hours before the vote: two supporters, one afternoon, two answers. The news nobody reported is operative paragraph 10, which urges states to explain projections’ “characteristics, limitations and appropriate applications” and to teach cartography critically in schools — Tissot’s 1859 move, made policy: you cannot remove the distortion, so draw it. A/80/L.104 was read in full from the UN’s own document server, the US explanation of vote quoted from the US Mission’s own publication, and the Guardian’s “superfluous” declined because the word appears nowhere in that statement and makes a substantive objection sound procedural.
A Landmark That Contains Nothing Is Still Exactly One Landmark
Star Stuff spent Tuesday on the landmark a screen reader uses to skip the nav, and the day is a case study in checks that agree with you. 130 of 159 pages had no <main> at all — not a half-finished migration but an inconsistency nothing was measuring, since pages edited on 30 August lacked it while pages from 12 August had it. The rule came from the twenty-nine that already did it right, and the migration refused to write any file whose content was not provably byte-identical either side of the two inserted tags. Two field guides had the opposite fault and were wrong before the pass began: each used <main> twice as a grid container, and two landmarks both called “main” is worse for a screen reader than none, because there is no way to tell which one is the page.
Then the pass failed its own stated rule on twenty-four pages and its own new check passed all of them. On the front page the landmark opened inside the masthead div and was closed by that div one line later — 0 characters, 0 children, 0 of 137 cards, measured live. It now holds 297,592 characters and every card. The twenty-two field guides had an existing grid container promoted into a landmark, so the landmark was one grid rather than the page: one guide held 2 of its 14 entries and 8% of the page, and is now at 81% with the h1 inside. And on the page carrying the site’s own signature phrase, the hero and the contents sat outside the landmark — which is also what the search index reads, so the phrase was unsearchable on the page about it.
So an eighth check, asking the property instead of the count: is the page’s own <h1> inside the landmark? One rule catches both shapes. Deliberately not a div-depth check, which was the obvious version and is wrong here — the zine template closes its shell before its main, so depth would have failed roughly 131 correct pages to catch 24 broken ones. Two of our own derivations were wrong on the way and both are recorded: matching <h1> in raw source counted hits inside <head> and reported twenty-five affected pages when the answer is twenty-four, and the DOM parser reported the field guides’ landmarks as empty because it does not run the JavaScript that builds those entries. A landmark that is also a grid container is a grid and not a page — now written into the site’s own instructions rather than re-derived, with the note that catching either failure shape needs a rendered page, which no gate here has.
An hour later, five pages turned out to have titles that looked like headings and were not — a cover title block, a playlist title, a broadside tagline, two title spans — and on two of them a screen reader’s headings rotor came back completely empty. All five now carry a real <h1>, and nothing moved, which was measured rather than assumed: computed position, width, height, font-size, font-weight and both margins captured before and after on all five, identical on every field. One trap caught before it shipped: joining two title spans onto one line inside the new <h1> made the accessible name read “NeurodiversityParadigm” — both spans are display: block, so the missing space is invisible on screen and exists only in the text a screen reader speaks. With the last of them fixed, the check stops counting and starts enforcing. The count existed so that it could shrink, and it shrank to zero.
The same week produced a seventh markup check, against a deliberately high bar: a fault must be invisible to every other gate and have recurred often enough to prove that remembering is not a control. Both were met — a card shipped inside another card in July and again this week, the second time with a written note about it already in front of us. It is depth-counted rather than string-matched, because a wrap contains a <details> with its own <div>, so “the next </div>” would be right only by luck; token-wise on class names, so card-wrapper does not answer to card-wrap; and regression-tested against the real broken file rather than only against synthetics. Six decoys confirmed silent, including the class inside a <style> selector, a <script> string and an HTML comment. 160 pages, 155,807 tags, 0 findings.
And on Saturday the limit of a hermetic gate got stated plainly. Ryan opened the new listening rack and two cards said “Video unavailable” — both had passed the collection’s own embed check the day they shipped, and both still pass it, because oEmbed confirms identity, not existence. All eight gates were blind and each was right to be. Fixing only the two he saw would have left the shelf in the state that produced them, so all 269 embeds across the Sound shelf were swept, turning up two more that had been live for weeks: a dead video on the Dolly rack and an age-restricted one on the Bowie rack — an age gate blocks embedded playback as completely as deletion while looking perfectly healthy from outside. The new checker was written and then deliberately not made a ninth gate, and the file argues its own case: all eight gates are hermetic and never touch the network, and a gate that can turn red while nobody has edited the repo is one people learn to push past.
Eight Zines, a Field Guide, and Three New Shelves
Already an Entity (No. 81) is built on a bibliographic fact. In 2020 Martuwarra RiverOfLife was first author of a paper in Transnational Environmental Law, reported as the first article in a law journal authored by a river; the same year a Challenges paper by eight Indigenous authors across four continents carried a dagger on its authorship line — “We formally acknowledge Mother Earth as our senior author, with her own presence and voice.” The method is deliberately inverted, and that inversion is the piece. Running our usual register on First Law would make physics or statute the validator of another jurisprudence and turn Aboriginal Law into an instrument for an argument of ours — so the object of study is settler law instead, and Mary Graham’s charge that ours is incomplete because destruction can be immoral and not illegal. Every person is attributed to a nation, on the page, in a table: three Aboriginal nations, three bodies of law, not a chorus. Graham is quoted from her own paper rather than through Yunkaporta, because letting one Aboriginal author be the conduit for another’s thought was the specific failure to avoid. Four sources read directly and open access; two cited and declared not read in full; and C. F. Black’s The Land Is the Source of the Law named on the page as a gap rather than used as a source. The rights-of-nature story then breaks on its own primaries — Poelina and Perdrisat record ambivalence about legal personhood, because the River is already an entity. It is the one zine in the series that ends without a claim about Neurodivergent belonging, and the last spread is the reason.
The Play You Keep (No. 82) reads Galpayage Dona et al. at the primary: bumblebees rolled wooden balls 910 times and gained nothing by it, all five of Burghardt’s criteria met, the place-preference result, the three-to-seven-day window. Every hedge the authors published is on spread 9 rather than tucked into the colophon. The re-scoping is worth recording: the proposal arrived as Helen Edgar’s The Weird You Keep, and its wyrd spine is her own earlier zine, with a companion already built for it — while the bowerbird broke its own argument, because a bower is courtship under sexual selection and therefore has a payoff, which is the one thing this zine’s play does not.
There Have Been So Many Earths (No. 84) is the collection’s second scroll zine and a nine-link chain where a section’s height is the time it covers — so Chicxulub is one screen you fall past, and the wait between oxygenic photosynthesis and the sky changing colour is three, mostly empty. Built from Hank Green’s video with two of its own citations corrected: the 3.7 Ga Isua stromatolites are disputed because some of the bumps point downward, and stromatolites do not grow down, and the famous thirty seconds is a 2017 documentary rather than a finding. A globe on every link shows the land at that moment, dashed before 1 Ga because that is as far back as a full-plate model reaches — a crisp coastline older than that would be an invention.
The Lizard That Was Never There (No. 85) is the collection’s longest chain at sixteen spreads, and the shape is the argument: what was told about the brain against what was observed in it. 1878, Broca names a rim of cortex le grand lobe limbique — limbus means border — and claims no function. 1937, Papez proposes a circuit for emotion, as a proposal. 1952, MacLean fuses the two, and a name for a place becomes a name for a job.* Then three brains stacked by evolutionary age, each supposedly independently conscious, and chronology pointed at living animals becomes a rank.* The hinge is Karten, 1969: the bird forebrain carries the same neuron types and the same wiring as mammalian neocortex, in nuclei rather than layers. The observed rail goes live in 1969; the told rail runs another fifty-one years without touching it. Spread 6 publishes our own implication rather than the flattering half** — Sagan is one of the two figures this site stands on, and his book is the principal vector for a model already contradicted when it won its Pulitzer. Reach is not a filter.* What replaced it is better than the story: 182 mammalian species, not three floors but two systems that trade, olfaction against neocortex at r = −0.63 — and the division is not age but the shape of the information. Sight, touch and hearing arrive with space in them and get maps; smell arrives with none and cannot. The armadillo spends its room on smell and is not a failed monkey.*
Older Than Either of Us (No. 86) and A Field Guide to Bird Minds (FG 23) went out together on nine bird-cognition papers, and the thing at the centre of them is an unresolved argument rather than a finding. The spine is Stacho et al. (Science 369, 2020): radial and tangential fibres crossing at right angles throughout the pigeon and barn owl forebrain, the same motif iterated across three senses and two species — a cortex-like canonical circuit with no cortex — with the paper’s own boundary printed alongside it. The drop carries all three positions rather than picking one: Stacho closes on an ancient microcircuit conserved from the last common stem amniote; Güntürkün, Pusch & Rose argue convergent evolution and call their own list speculative; Zaremba et al. find conserved inhibitory cell types alongside transcriptomic convergence. “Convergent evolution” is the version that travels. The belonging claim survives either answer, which is how you can tell it was never resting on the ranking. Three refusals came with it, and the first is the most useful: the obvious zine was declined as three times already spent, found by grepping the argument rather than the topic.
Two Peppers (No. 87) went back to the accepted manuscript because the review chapter had dropped the inconvenient half. Chiltepin, the wild Capsicum annuum: +33% carbon and three times the fruit. Jalapeño, the domesticated plant of the same species: −11% carbon, but +157% water-use efficiency and equal fruit on 65% less water. Cherry tomato doubled, and the panels ran about 8.9 °C cooler over the crops. The two papers in the drop contradict each other and the contradiction is the argument: tomato doubles under Arizona panels and loses a quarter under mild Michigan shade, because what was scarce differed. Shade tolerance is a property of the plant and the place, not a number you can carry between them.
The Name Was Already the Verdict (No. 88) spun out of the same day’s listening rack. Stations banned Link Wray’s Rumble “for its incendiary title” — a record with no words in it — and the record was made as “Oddball.” The word that got it banned was applied afterwards by a listener, and the two best accounts of who that listener was name different people; tagged contested and left there, because both agree the name came from the listening side and was then treated as content. Corrected against the musician: the New York Times and Rolling Stone both put a pencil through the speaker cone, a sentence that has travelled everywhere including into our own first draft — Wray says a pen, through the tweeters, and that he left the big speaker alone. And the joint the chain would most like to have is refused outright: no documented line runs from the 1954 Kefauver hearings to a programme director in Boston, so that link is contested and the spread says deleting it leaves the chain standing. A Burke chain is built to feel inevitable; the tags are what stop it.
Alongside them, We Were Punk First gathers thirty-eight recordings of Indigenous punk, metal and hard rock with Gregg Deal and Dead Pioneers as the spine, built from Al Carroll’s Indigenous Punk read in full — and the first thing it did was correct its own source twice, on the page. Link Wray entered the Rock and Roll Hall of Fame in 2023, in the Musical Influence category — not 2007, and eighteen years after his death rather than two. The guitarist is Josh Rivera, not “Gregg Rivera.” Three further claims were declined and the declining printed too, including leaving the positive account of powwow drumming to the people who play it. It is also the first rack here that argues with another rack: It Take a Joyful Sound holds that naming a thing helped it happen; this one holds that naming is not founding.
And three shelves opened. Glimmers is the fourteenth collection and Ryan’s call — a shelf rather than a loosening, because a zine takes one settled fact and follows it until the belonging claim is already inside it, and settled facts can wait. Its axis is occasion, a fourth kind of sort here, and three things follow, all of them constraints rather than permissions: entries are dated rather than numbered; both the event date and the publication date appear on every entry, because a piece written close to an event was written with less than the full picture; and it runs the same eight gates, the same ledger row and the same changelog entry as the slow work. It is named for a word this site removed from a zine two days earlier — No. 78 was repointed because the frame outlived its foundation. A frame is load-bearing; a label is not.* Glimmer Wire became the fifteenth on Friday, on a fifth sorting axis — issue.* And Triggers is the sixteenth, opened for bad news that comes with a way out, sorting by form — the third collection that does, and it says so rather than inventing a sixth axis. Its convention is a five-tier readiness grade on every response, because a grade is about deployment, not merit: afforestation is deployed and nearly useless at scale, while net-negative CO₂ is merely proposed and is what every “we can return to 1.5” sentence depends on.** Overshoot gives five mechanisms for why 1.5 °C is not a small number — a global mean over a planet 71% ocean, so land was already at 1.59 °C when the globe read 1.09 °C; a 20-year running average, so 2024 was not a crossing; the once-in-fifty-years hot day 8.6× more frequent and 2.0 °C hotter; the last ice age 6.1 °C colder, so 1.5 is a quarter of that distance travelled in 150 years; and the surface figure as a receipt for 381 zettajoules of which about 1% stayed in the air. Current policies give 2.6 °C by 2100, which makes 1.8 °C the optimistic case rather than the forecast. And its mobility entry corrects the question it was written to answer: Pacific Islanders have explicitly rejected the framing of “climate refugees,” and the IPCC finds with high agreement that agency over the move decides whether it helps or harms.
The Digest That Publishes What It Declined
Glimmer Wire is the graded scan behind Glimmers, published with its grades showing instead of kept in a drafts folder. The first edition went out with 7 filed, 3 held, 5 leads, and zero graded VERIFIED, because six publishers refused a direct read on the compile date — so every item is sourced to institutional releases plus publisher metadata and DOIs, and each entry names what was read and what was not, as a standing notice above the fold rather than a footnote. Items sort by lag: how long a finding had been public before a news feed carried it. Median 36 days, exactly one same-day, two over a hundred days old — which is the argument for weekly rather than daily.
The section that earns the page is Held. The changelog records what we got wrong and fixed; the ledger records what we checked; neither has ever recorded the story we declined to tell. Edition one declines a humpback “super-group” claim that the peer-reviewed survey of the same water contradicts.
Then edition two published a whole edition that could not open a single source and said so on its own face. Every publisher, aggregator, press wire and DOI resolver the scan reached for was refused at the egress proxy before the request arrived — science.org, sciencedaily.com, phys.org, eurekalert.org, sciencenews.org, arxiv.org, api.crossref.org, doi.org. That is one step further out than edition one, which at least reached institutional press releases, so the whole edition rested on search-result summaries — leads, never citations — and nothing in it was allowed to be promoted to a zine, a Glimmers entry or a card. When the block lifted hours later, every source was tried again, and what the primaries mostly bought was not confirmation. One upgrade. Five stayed held, each with the exact refusal named — PNAS 403, Science paywall, Cell 403, Annual Reviews 403, and Nature Materials at $39.95, where the authors’ own preprint was read in full and the item still not graded verified, because a preprint is the authors’ text and not the version of record. Four more entries lost claims the papers do not make. And the two corrections that cost the most were self-inflicted — the misread title and the anomalous Hall effect, both above. The lag axis was left honest rather than redefined mid-archive: two items were public as preprints 753 and 478 days before the news carried them, and that is named in their entries.
One item did make the trip. Until There Were Ten published the day the paper appeared and became the first item on the site to move from plausible to verified by getting hold of the primary. Saturn’s hexagon has been in every image since Voyager, a lifetime exceeding 44 years with no counterexample anywhere on the planet — and the southern hemisphere, tipped out of view from mid-2012, came back in 2023 carrying a ten-sided wave at 58°S to 63°S. The piece is built on the paper’s refusal of the tidy reading: the decagon is “not uniform in longitude”, its corners oscillate on a mean 32.5-day period and the wobble travels, and it may be transient where the hexagon is robust. One example, held still for forty-four years, had looked exactly like a law.
Two editions on one page was already too much, so the split came on Friday — and immediately exposed an instruction that would have undone it. Nothing was re-graded or re-worded: both editions keep their text, grades, amendments and anchor ids verbatim, including edition two’s asymmetric id suffixes, which look inconsistent and are load-bearing, because two external deep links point at them and a Netlify 301 does not carry a fragment. Then the audit found the trap: the old single page’s source comment — “One page, every edition kept… New editions are PREPENDED” — had been copied verbatim into both new files, sitting in the source of the very pages it misdescribes, where the next person or agent to read it would have prepended edition three into edition two and recreated exactly the problem the split fixed. It was replaced with a comment recording that the old instruction was wrong, so meeting a surviving copy elsewhere is recognisable as stale.
Two Sheets, and the First Statute on One
Monotropism is a new Why Sheet and the eighth broadside, magenta and violet, written because Fergus Murray and Helen Edgar asked us to turn their booklet into one — the moment being a meeting where a young person is called defiant, disengaged, or obsessive. The attribution was the part that had to be right, not the prose. The six starting points are Fergus Murray’s, from ‘Starting Points for Understanding Autism’, January 2019, republished by Thinking Person’s Guide to Autism, carried in three places across the two sheets because Helen asked specifically that the source be named. The theory is Dinah Murray, Mike Lesser and Wenn Lawson’s, 2005, and the sheets say so. Fewer things at once, more attention on each — inside the attention tunnel there is depth and detail, outside it things get missed, and moving the tunnel costs energy, which is most of the distress attributed to Autistic people rather than a fault in them. The Monotropism Questionnaire caveat is load-bearing and stays on both sheets: it went viral as an autism assessment and its own authors objected in public, so a sheet that signposts it without saying not a diagnostic test makes the problem worse. The licence took a decision rather than a default — the sheets are CC0 like the rest of the line, and CC0 cannot require attribution, so the credit is asked for on the page instead, which is the one arrangement that stays honest on work adapted from somebody else’s. Two citations were wrong until the sources were actually opened: Sonny Hallett’s Loops of Concern is 2021, not 2018, and Julia Leatherland’s is a deposited doctoral thesis with a longer title than we gave it — both fixed on the live page within the hour, because a wrong year in a reference is the kind of error nobody reports and everybody inherits. A third check ran the other way and corrected us: the sheet list on /why/ was flagged as missing a sheet, and it was not — that list is generated from child pages and our local mirror was stale. Check the live page before filing a bug against it.
Masking and Burnout and its broadside — the ninth, with a Markdown copy in the open repo — exist because a community member’s son was refused a special education evaluation on the grounds that he is doing fine. He is high-masking, he is at grade level, and his 504 had already failed to prevent an Autistic burnout that ended his access to school for the second half of fifth grade — which his mother had put in writing, in the request, before the observations began. The refusal came back citing the grades. Grades measure output. Access is about cost — and a school that measures only output keeps finding no need until the child can no longer attend, at which point the record of coping becomes the reason no help arrived. Side A is that cycle; side B is what to say, what to ask, and the two federal regulations that already name this exact situation: child find covers children “even though they are advancing from grade to grade,” and FAPE is owed “even though the child has not failed or been retained.” It is the first sheet to carry statute, and that is not a new house habit — the refusal it answers is a written administrative decision whose stated reason the regulations forbid, so quoting them is the part that changes what happens in the room. Most subjects have no statute to point at. Built on Helen Edgar’s Supporting pupils through Autistic Burnout, used with her permission, her words attributed and outside our CC0 grant. Two findings stayed off the sheet on purpose: the camouflaging-and-suicidality research is a small correlational study of undergraduates and the children’s burnout audit is a twenty-child conference abstract, and a sheet has no room for the hedges either one needs. A page that says “masking is linked to suicide” without them is worse than one that says nothing. Both are on the Why Sheet, hedges visible. We also wrote the family a response letter to send the school — not published, and not in this voice.
The Bylaws Before the Vote
A consultant sent us one conflict of interest policy, and the accountability hub came out the other side larger by most of a shelf. The sample was sound and its central mechanism was illegal for us* — it authorizes a conflicted transaction by vote of the disinterested directors, which is exactly the §4958 rebuttable presumption a public charity gets and precisely what a private foundation does not: under §4941 self-dealing is flatly prohibited, no vote cures it, and the penalty is strict liability.* So the rewritten policy carries both regimes at once, saying plainly which governs today and which starts the day our conversion to public charity is approved — because a policy that quietly assumes the status you are applying for is a policy that is wrong right now. The Whistleblower policy was rebuilt from our 2022 formation documents with the Covenant named as the code of conduct it enforces, and then the consultant read it and found the real gap: it explained every way to bring a concern to us and said nothing at all about the right to skip us entirely.** It now names the IRS, the Delaware and Texas attorneys general, and law enforcement, states that no agreement we ask anyone to sign can take that right away, and cites the federal crime that retaliation for an external report actually is. A policy that only explains how to come to us is a policy about us.
Then the bylaws, published for the first time — all forty-two sections of the document executed in 2022, with one sentence deleted from §2.1: the clause requiring every director to be a member of the founder’s family, by blood or marriage. There have been no family members on our board since the founder’s spouse resigned as part of the diversification behind our public charity conversion, so the governing document had been describing an organization we are not. A rule you have to keep making exceptions to is a rule that should be changed. The page is marked Proposed, not adopted, because the board does not vote until this month and the 2022 text governs until it does — and rather than ask anyone to diff two documents, the page quotes the removed sentence in full. We also reproduced a mistake rather than tidying it: the executed document labels its Miscellaneous article “ARTICLE VI,” duplicating the Amendments article, while its own sections run 7.1 to 7.6, so the published page carries the error, explains it, and flags fixing the heading as separate housekeeping. A published version that quietly differs from the signed one is not a cleaner document, it is a second document.
The Conflict of Interest Statement is now a working 51-field form — signature capture, a date picker, conditional logic so each “describe” box appears only when someone answers yes — and it is public over our own recommendation to keep it private. That recommendation was wrong, and Ryan said so. It conflated two unrelated things: a public page exposes the questions, not the answers, and the questions were already published almost line for line in the policy. Hiding the form protected nothing. What it cost was real — most small nonprofits needing one of these have to buy a template or pay a consultant for the sample, and the parts worth copying are exactly the parts templates get wrong: disclosing is not an accusation, the circle is wide but you never have to name anyone, there is another way in if forms are a barrier, and it asks about divided loyalty rather than about your capacity, your health, or your care responsibilities. So the page leads with the sentence a compliance form on an open URL actually needs — this page is public, your answers are not — and invites reuse.
Board Roles & Responsibilities, a signable Board Member Agreement and Board Member Onboarding went up together, and none of it came from a template — our consultant’s sample packs and the BDO policy set have no board agreement in them at all, so this was built from our own bylaws. The half worth stealing is the list of things we refuse to ask for: no give-or-get, no personal donation, no fundraising quota, no wealthy contacts, no travel, no unpaid professional labour, and no being non-Disabled or performing wellness. Most board descriptions quietly select for wealth, free time, and good health, and ours says so in as many words rather than letting a Disabled candidate work it out from the silence. The agreement runs in both directions: Part 4 is what the Foundation commits to the director — indemnification under Article V including advancing the cost of defending a proceeding before it concludes, papers early enough to be read, reimbursement, and accommodations without a diagnosis or a negotiation — and the incoming director ticks a box confirming they read that half, because a promise nobody acknowledged receiving is a promise nobody can hold you to.* The onboarding page opens with If You Only Read Three Things, and one of the three is Where the Money Went: you should know we spend far more than we take in before you join, not after.* It also names the tension we cannot design away — we will flex almost everything about how you participate, and we cannot flex whether the board can meet, because quorum is law and §2.8 permits acting without a meeting only by unanimous consent. Onboarding is our obligation, not their homework.**
Friday evening added the half nobody had written yet. Staff & Volunteer Onboarding is deliberately not a mirror of the board page — it routes to the shared policies rather than restating them, and says plainly that the bylaws and board pages are not staff’s job. Its centre is the duty only staff can discharge: our retention policy publicly promises we delete what mutual aid applicants tell us, and the Board can adopt a policy but cannot delete a form submission. Writing it surfaced the gap: Article V indemnifies directors and officers, not volunteers, so our unpaid Co-Creative Directors may have no protection — §5.7 permits extending it by resolution, and that is now an eleventh board decision on the log. Then the Volunteer Agreement, which leads on copyright because that is the question volunteers most often have no answer to: a volunteer is not an employee, so work-for-hire does not apply automatically and ownership goes undocumented by default — which hurts the maker, not the organization.* Non-exclusive licence, writing under CC BY-SA, art rights stay with the artist, credited by the name they choose. And it says plainly that Article V does not cover volunteers and that we will not claim otherwise until the Board acts.* The three signable forms then got document IDs of their own, which the policy pages had and the forms did not — backwards, since the forms are the instruments people actually sign and that we retain for seven years — each footer carrying one line that makes the ID do work: “If you have signed this form, the version above is the one you agreed to.”**
And the honest entries are the ones about ourselves. The Transparency Log moved off /now/ to a page of its own and grew past thirty entries, including the summer board meeting we missed, the vacant Treasurer seat, and financial figures on Frequently Requested Information that still showed numbers we had already publicly corrected in August. The worst error of the week was ours and it came from reading our own website: no page named a President, so we wrote “Board Chair” into the routing of two policies — and the bylaws make President and Secretary the two required offices. That error had reached the September board agenda, the officer discussion, and the reasoning about who signs checks. Fixed across every page and every routing line, then logged publicly rather than swapped in silence. We built a retention policy from a consultant’s sample and only afterwards found counsel had written us one in 2022, sitting unread in the formation binder; the two disagree about who is responsible, so the board decides and the 2022 version stands until it does. That went in the log alongside the wins, because a log that only records the things that flatter us is marketing.* The retention policy itself names two duties, not one* — accountability needs records, privacy needs them gone — and splits grant files accordingly: we keep the transaction and delete the story.** Keeping the right records is accountability. Deleting the rest is care.
A Footer, a Print Stylesheet, and a Palette That Was Right About the Wrong Ground
Ryan asked whether the stimpunks.org footer is sufficiently distinguished from the content above it. It was not — and the background colour was the wrong suspect. The band sits one Solarized step off the page, #073642 on #002b36: 1.15:1, a lightness difference of ΔL* 4.8, the tightest pair in the whole palette, designed as a background highlight inside content and never as a region boundary. What actually said “this is just another section” was everything else. The footer’s h2 was pixel-identical to every content h2; the hrule meant to demarcate the footer was the same 2px #808080 as the eight other separators on Start Here, and it rendered above the coloured band rather than as its edge. Four rounds fixed it — heading step-down, 81px of padding top and bottom, the rule reborn as a full-bleed border-top on the band itself, the redundant separator gone. The lesson is that full-bleed versus inset did the work, not the colour of the rule: edge-to-edge against inset is a categorical difference, and light grey against slightly-lighter grey is not.
Then the real finding surfaced, and it was never a footer bug. Two theme slots resolve to base02, and every panel built on it drops its text below WCAG AA. On the front page, callout body text runs 4.11:1 and links inside those callouts 4.06:1 — the worst pair on the site — pullquotes 4.11, block quotes 4.11 on base02 while the same quote block passes at 4.75 on base03. All eight Solarized accents were checked against both grounds: none clear 4.5:1 on base02, and exactly three clear it on base03 — yellow, cyan, green, which are precisely the three we use. The palette was chosen correctly. base02 is what broke it.* And a measurement trap worth carrying forward: every font size we read was a clamp() floor.* The theme is fluid, typing “24px” in the block editor sets the clamp’s maximum, and the footer heading only reaches it at 762px wide — 18.7px on a phone.** WCAG’s large-text threshold is 24px, so a fluid heading changes accessibility class with the viewport. There is no size-based route to conformance. The colours have to be fixed.
Separately, Ryan asked whether the three print settings he picks by hand could be pre-selected. No browser lets a page touch the print dialog — there is no API and there never has been, which is correct, because a site that could decide what your printer does has too much power over your machine. That turned out not to be the problem. stimpunks.org carried one @media print rule in 211KB of CSS, and it belonged to Jetpack search. Printing a Why Sheet meant hand-picking Firefox’s Simplified format on every job — a Firefox-only feature that also throws away the font and size the reader chose in the Aa Reading panel. Worse, dark mode was reaching paper: the palette pins color-scheme: dark as an author declaration so it survives into print, and with Background graphics off — the browser default — Solarized base0 landed on white at 3.16:1. Firefox forces a light scheme for print and hid it; Chrome does not, so the person who prints these most was the one person who could not see it. One rule fixes it. Then a Print or save as PDF button in the reading panel, generalised from the trick the broadsides already use — content only, the reader’s own font and size kept, and our footer carrying the title, URL and date. Printing Pages from Stimpunks.org covers the one setting still theirs.
And the footer got a garden in it. The reading options block is now bracketed by two 100px decorative strips — a rhizome above, a waterline with lily pads below — drawn from things the site already publishes rather than from stock foliage. The Autistic rhizome is defined here as a network with “no hierarchy or dependence on another’s existence,” so the art has no trunk and no centre — which is also the only reason it can tile, because a composition with a focal point cannot repeat. The lily pads carry the six documented pad tints, so in dark mode they collapse to the single panel tone exactly as the page’s real lily pads do. The art inherits the behaviour instead of imitating it, from one file containing no hex values at all. The finding worth keeping is the boring one: whether inline <svg> survives WordPress’s KSES filter in a Custom HTML block was genuinely unknown — the Aa Reading panel builds its controls in JavaScript specifically* because form elements do not survive it, and there was no precedent anywhere on the site.* It survives, and that is now written down with the one-line re-check and the fallback, because the failure mode is a block that renders nothing and says nothing. One trap was caught by walking into it: SVG paint-server ids are document-global and url(#id) takes the first match, so two strips sharing a pattern id means the second silently draws the first one’s art and inherits its colours** — which briefly convinced us the dark-mode collapse was not working.
The Knowledge Garden
The team already had every file and still could not use any of it, so we built the way in and named it the Stimpunks Knowledge Garden. The Drive mirror had been publishing since August; what was missing was an interface, and the first proof of that was Ryan being unable to find the folder at all — the setup doc named a shared drive that does not exist, so every teammate would have hunted exactly as he did. The mirror became a team product rather than a copy of a disk: out went 554 MB of video no text tool can read and 13,847 machine JSON records, in came the extracted text of all 371 readable assets — because 314 of our 457 originals are PDFs, and a PDF is a wall to a search tool. Then a latent bug: rclone sync applies its filters to both sides, so tightening a filter hides files rather than deleting them. Drive still held all 33,179 objects while the sync reported success. The garden itself is 15,920 pages built from six corpora, searchable across all of them, faceted by our own glossary concepts — which is the whole reason it is a custom build and not a theme, since no theme can read a concept index we already had. Password-gated, and the deploy tool treats that as the check that matters: it fetches the live site afterwards and calls a 200 a failure even when the deploy succeeded, because the password is a dashboard setting nothing in the repo can enforce.
Later in the week the mirror caught something worse than a missing file. Twenty-three field guides had been mirroring their whole argument and none of their entries — the guides build their cards in the browser, so a static fetch came away with the prose and nothing underneath it. A Field Guide to Bird Minds mirrored 8,223 characters of real argument while Psittacus erithacus, Corvus moneduloides and eleven of its thirteen entry names were simply absent. The recovery needed no new machinery: Star Stuff already loads every page in headless Chrome and commits what it read, for its own search, so the mirror now reads that same index rather than taking a second guess at the HTML. 167 pages, 23 augmented, 312 entries recovered, the other 144 byte-identical. An index that cannot see the writing cannot connect it to the reading.* Every failure that week looked fine until a person tried the actual path.* The folder nobody could find. The search box that vanished because three builds deleted an index. A _headers file the host rejected outright, so noindex silently never applied. A token that authenticated perfectly, reported an 18 TiB quota, and listed absolutely nothing, because it belonged to the wrong account. Broken systems, not broken people — including the ones we build ourselves.**
Mission, and a Table of Falsifiers
Foundations had a page for what it holds, what it owes, how it is made and how it connects — and none for what we are trying to do. Mission is that page: eleven aims in four groups, each linking the page where it is already being practised. Belonging owed rather than earned, and variation as the rule at every scale. One settled fact followed honestly, every quotation traced to a primary, the checking published errors first, and science celebrated by being practised rather than by having its authority defended. Hopeful news reported better than the coverage of it, the weekly scan published with what we declined, and no harm named without a graded ladder. Work done in the open, readable and printable and free — plus the AI experiment stated rather than smuggled.
It closes on a table of falsifiers: eight observations a reader could make from outside that would show an aim had quietly been dropped. Four have already happened, and they are named on the page. The second half reads the Stimpunks mission sentence into its three verbs, puts this site under each, and then marks the limit — Stimpunks’ first sentence is about direct support and mutual aid, and no zine here has paid anybody’s rent.
The same day, the site learned to answer “what came out this week” without anybody maintaining the answer by hand. What’s New lists every piece newest first with the title and the line it leads with, and the RSS feed carries the same list — neither is hand-written. Titles and taglines are lifted from each piece’s own card, the same map the markup checker already derives badges from, and dates come from git itself. Easter Eggs are excluded, because a listing announces and a listed egg is not off the path. One date override was needed, and the count of overrides prints on every run rather than living silently in the code.
Week 36 Changelog
- Published the Week 35 changelog and the Week 35 reflection — the second recording what the week cost rather than what shipped, including four attribution errors found on our own pages on a Friday and still unrepaired on the Sunday. Finding an error and repairing it are separate days’ work.
- Published Monotropism as a Why Sheet and the eighth broadside, at Fergus Murray and Helen Edgar’s request, adapted from their 2026 booklet — six starting points credited to Fergus Murray’s 2019 essay in three places, the theory credited to Dinah Murray, Mike Lesser and Wenn Lawson, and CC0 with the attribution asked for on the page because CC0 cannot require it.
- Published Masking and Burnout and the ninth broadside, plus a Markdown copy in the open repo — the first sheet to carry statute, because the refusal it answers is an administrative decision whose stated reason child find and FAPE forbid.
- Published Campfire Learn Together: The Subtle Art of Normalising Dystopia for Sunday 6 September, paired with our own Boring Technology page — same word, opposite directions. Its synthesis extends the film to the sentence it never reaches: administrative burden is DRM for human bodies. Three pushbacks, the load-bearing one being that masking is not self-censorship and is not complicity. A fabricated archive.org URL was caught and removed before publishing — a constructed URL is an invented citation wearing a hyperlink.
- Published our bylaws for the first time, marked Proposed, with the founder’s-family clause deleted from §2.1 and quoted in full on the page, and the executed document’s article-numbering error reproduced and explained rather than tidied.
- Rewrote the Conflict of Interest policy to carry both §4941 and §4958 with an explicit sequencing section, and published the Conflict of Interest Statement as a public 51-field form over our own recommendation to hide it.
- Rebuilt Reporting Concerns from our formation documents and then added the right to skip us entirely — the IRS, two attorneys general, law enforcement, and the federal crime retaliation would be.
- Published Board Roles & Responsibilities, the Board Member Agreement, Board Member Onboarding, Staff & Volunteer Onboarding and the Volunteer Agreement — none of it from a template, all of it from our own bylaws, and Part 4 of each agreement is what we owe the signer.
- Published a Document Retention & Destruction Policy, gave the Transparency Log a page of its own, corrected our officer titles sitewide after publishing “Board Chair” for an office the bylaws call President, and gave the three signable forms document IDs of their own.
- Published Printing Pages from Stimpunks.org, added a Print or save as PDF button to the reading panel, and stopped dark mode reaching paper at 3.16:1.
- Fixed the footer’s distinction from content in four rounds, and found the real problem underneath it: every panel built on base02 drops its text below AA, links inside front-page callouts at 4.06:1. Added a rhizome and a waterline to the footer, and proved inline SVG survives WordPress’s filter.
- Built stimpunks.garden — 15,920 pages from six corpora, faceted by our own glossary, password-gated, with the deploy tool treating a 200 as a failure.
- ★ stuff published eight zines — Already an Entity (81), The Play You Keep (82), The Width of Itself (83), There Have Been So Many Earths (84), The Lizard That Was Never There (85), Older Than Either of Us (86), Two Peppers (87) and The Name Was Already the Verdict (88) — plus A Field Guide to Bird Minds (FG 23) and the listening rack We Were Punk First.
- Added three Foundations pieces: Rainbow Serpents and the Laws of Thermodynamics, Six Principles and a Room and Mission, which closes on eight falsifiers, four of which have already happened.
- Opened three collections — Glimmers (fourteenth, sorted by occasion), Glimmer Wire (fifteenth, sorted by issue) and Triggers (sixteenth, sorted by form, every response carrying a five-tier readiness grade).
- Published five Glimmers entries and two Glimmer Wire editions, the second of which could not open a single source and said so on its own face — and Until There Were Ten became the first item on the site to move from plausible to verified by getting hold of the primary.
- Gave all 159 Star Stuff pages exactly one
<main>landmark, then found twenty-four of them pointing at nothing and added an eighth check asking whether the page’s own<h1>is inside it. Added a seventh check so a card cannot ship inside another card. Gave five pages a real<h1>and turned the count into a rule. - Swept 269 embeds across the Sound shelf after two dead videos passed the collection’s own check, found two more, and deliberately did not make the new checker a ninth gate — a gate that can turn red while nobody has edited the repo is one people learn to push past.
- Recovered 312 field guide entries into the site mirror by reading Star Stuff’s own search index instead of guessing at its HTML, with a gate that separates client-rendered pages from server-rendered ones with no overlap at all.
- Retracted a bug we had announced: the
<br>accessible-name fault does not exist in Chrome, and the 347-edit sweep was written, measured, and discarded rather than shipped. - Corrected ourselves in public on the fibromyalgia entry, the Kanzi card, the corvid mirror test, the elephant species, the anomalous Hall effect, and Which Distortion‘s own hero line — and parked the piece the first of those was going to become, with exactly what would unpark it recorded.
- Added a mass-deletion guard to the nightly
notes/backup, which our own knowledge system documentation had claimed for both commit paths when only one had it. It withholds rather than aborting, because a timer that refuses is a day with no backup and nobody watching — and it counts deletions after rename detection, since a reorganisation moved 1,307 files that afternoon. A doc that promises a check nobody wired up is worse than no check, because you stop looking.
This week the same fault kept turning up in six different materials: a condition that was printed right there in the source, attached to the claim, and came off somewhere between the paper and the headline. Entropy in an isolated system. Solar with low-cost financing. Seventy micrometres. A projection more accurate about sizes, where sizes are critical. The literature was more careful than the coverage every single time — which is a finding about transmission, not about scholarship, and it is ours to fix too, because four of the week’s sharpest corrections were of our own pages. The checks told the same story from the other side. A landmark that contains nothing is still exactly one landmark. A build that reports 100% coverage on a path it skipped. An oEmbed that returns 200 for a video that will not play. Every one of those checks was measuring something real and none of it was the thing that mattered, so the fixes were not more gates but gates that ask a different question. And on this site we put the governing documents on the open web — the bylaws before the vote rather than after it, the compliance form public over our own advice to hide it, the missed meeting and the vacant seat and the policy counsel had already written us sitting unread in a binder, all of it on a log next to the wins. Broken systems, not broken people — and a system you cannot read is one you cannot hold to anything.