I spent the week working on Star Stuff with Helen and hanging in Discord with community. It was a nice rhythm.
Week 34 runs Sunday 2026-08-16 through Saturday 2026-08-22. The day headings below carry their dates.
Best Accomplishments
Creating the “Props & Gratitude” channel in Discord, which the community promptly started using.
Theme
Props and gratitude and attribution.
Sunday — 2026-08-16
Accomplishments
- The whole Sunday rhythm ran end to end for the first time, and Week 33 closed complete on both documents. The Week 33 changelog was drafted from a running file carrying 55 bullets across all seven days — the first week with no gap to reconstruct — then published, then run through the post-publication checklist: 109 links verified live, category and tag confirmed, featured image present, Bluesky and Discord drafts written. The reflection was audited straight after and needed almost nothing: no broken links, no repo-relative links, no internal-system links, no empty days, and fourteen first-person passages every one of which was legitimate — quotations, titles, regex literals and template headings — so none were touched. Two additions to its intro: the companion link to the published changelog, and the note that our repo links answer 404 rather than 403 for readers without staff access. That note is Week 33’s own Sunday insight applied to Week 33. Week 32 shipped without it.
- Two zines, two field guides, and four attribution corrections on ★ stuff — both new pieces grown from our own most-read pages. How Anything Gets Built (No. 54) sets the five love locutions beside five mechanisms by which matter is actually assembled, and then refuses its own tidiness on spread 7: the physics does not justify the love. What Looks Like Chaos (No. 55) takes the textbook instinct claim about nest building and finds somebody had measured it — not one behaviour repeatable within a male weaver, personal handedness with no species handedness underneath, and roost styles that survive weather, kinship and immigration. A tradition, not a template. Their companions, The Ways of Being Alone (Field Guide No. 14) and Nests (No. 15), each found the word they were named for was wrong about the animal: a lone wolf is a disperser on its way somewhere, and most nests are not built.
Insights and Takeaways
- The post about claims made from memory contained two claims made from memory, and Ryan caught both of them. The first was a section header: “Our Own Numbers, Wrong in the Flattering Direction” — sitting directly above its own sentence saying we had been marking ourselves down by more than twenty points. The running changelog had it right and plain: the numbers turned out to be wrong in both directions. The draft compressed that into the more quotable half. The revenue figure flattered us; the program share sold us short; we overstated what came in and understated what we do with it, and only one of those made it into the heading. The second was structural. The Week 33 Changelog section shipped at twenty items because twenty felt like the right density — nobody checked. Measured against the last two published posts, the practice is near one published bullet per logged bullet at about seven hundred characters: Week 32 carried 34 of its 39. Week 33’s draft carried 20 of 55, at less than half the length. Both faults were fixed the same way, by measuring instead of asserting — which is the move the post spends five thousand words recommending. A standard you can state and not apply to the sentence stating it is not yet a standard.
- A guard refused the correct call while silently permitting the wrong one, and only the annoying half was visible. Logging Saturday’s work on a Sunday hit a refusal:
--day 2026-08-15 is outside this changelog week (2026-08-16..2026-08-22). The message was confident, specific, and wrong — it validated the day against today’s calendar week while writing to the file named by the week counter, and those two disagree every Sunday, because the counter only advances insidedraft-weekly-changelog. So the one morning of the week when the previous day still needs logging is precisely the morning the tool cannot do it. The dangerous half was the half that did not complain. A weekday name was never refused at all:--day saturdayresolved to the following Saturday and would have written a## Saturday — 2026-08-22heading into the Week 33 file — an authoritative-looking wrong date, in the one structure that exists so a missing day is visible. A guard that refuses a good call costs you a minute. A guard that permits a bad one costs you the record, and the first kind is the only kind you find by using the tool. The fix reads the window from the day headings already in the running file, which answers the objection the original code had written down against using the counter: it is the file’s own record rather than an invented mapping. Same fault, same fix, twice —changelog.pyand thenreflection.py, which had inherited it by sharing the function. - We spent a month building tools to stop trusting aggregators, and then found the aggregator was us. Every route we could follow for the five love locutions led back to our own page: Helen Edgar’s post attributes penguin pebbling to Amythest Schaber with an explicit “as far as I know”, her practitioner guide cites Stimpunks (2022), and Stimpunks (2022) cites Schaber. A closed loop with no primary at the bottom of it, and the hedge stripped somewhere on the way round. That is the exact failure Too Good to Check exists to catalogue, committed by our most-read page — and the reason it survived is the reason five copies of the misquoted Eno definition survived last week: repetition reads as corroboration, and our own material repeats loudest. The primary was never even missing. It was in the blockquote our own 2022 page already carried. A secondary source that quotes a primary usually links it, and reading the page beats searching around it. The second correction is the one to carry forward, because it is structural rather than careless: Field Guide No. 14 credited “the Stimpunks page” five times for an essay with Betsy Selvam’s byline on it the whole time. Every other citation on that guide is a named human. The one source that got flattened into an institution is the one that came from inside, and it got flattened precisely because the byline felt like ours. “Ours” is not a citation — and a collectively-run site is exactly where an author is likeliest to disappear into the masthead, on a guide whose own subject is that being alone is a practice with a name and an author rather than an absence.
Gratitude
- The three people whose names Sunday put back where they belonged. Betsy Selvam, whose Lone Wolfing: The Joys of Autistic Solitude is the whole spine of Field Guide No. 14 — including the four conditions it is structured on, which were quoted in her essay rather than found independently. She was already right about the practice; the wolves only turned out to agree. Amythest Schaber, who originated the five neurodivergent love languages that our own page put into currency and then, for a while, obscured. And Lindsey (brainsandspoons), through whom penguin pebbling reached us later — a name that only exists in the record because somebody went back and reconstructed the chain instead of accepting the tidy version where one person named all five.
- David Thornburg, whose fourth space came back because somebody opened the paper. Caves, campfires and watering holes are his, from Campfires in Cyberspace — and reading him at primary while fixing the credit turned up something the version in general circulation quietly drops: he names four spaces, not three. Campfires for information, watering holes for conversation, caves for concept, and life, for context. A whole space, missing from every retelling including ours, recovered as a side effect of correcting an attribution. And Tanya Adkin, whose name is first on both pieces that set out lilipadding and who had been dropped from a draft that assumed a sole author from a shared subject. Two names restored, and one idea restored with them.
Monday — 2026-08-17
Accomplishments
- Four pieces on ★ stuff, a broadside and a glossary entry here, and an arc that opened five days earlier closed. The Noise Floor (No. 56) is the ninth chain and the last of the four proposed on 13 August to be built — the other three shipped as Nos. 43, 44 and 50, and the one deliberately withheld is still withheld. Starlight joined Foundations as the piece that says out loud what the other seven presuppose about process, and with it the first plain statement anywhere on the site that generative AI facilitates the work — true since July and written down nowhere, which is a poor showing for a project whose argument is that the working should be visible. The Shell Was Your Calcium (No. 57) and A Field Guide to Eggs (No. 16) came as a pair. And on this site, Developmental Pace became the sixth broadside and then commissioned its own parent page — the first time a sheet has called a glossary entry into existence rather than compressing one that already existed.
- The knowledge system fed a zine the same morning it ingested the sources, which is the first time that loop has closed visibly. Fourteen files went into the library before seven in the morning — a profound-autism critique, scripted curricula, acoustics — and by the evening two of them were load-bearing in The Noise Floor. One of them changed the argument rather than decorating it: a 2026 thesis that put the zine’s central question to 311 noise-sensitive people and found “no habituation, but accumulation effect” — which breaks the analogy the chain was built on, because thermal noise is stationary and the measured human pattern is fatigue that builds. A resistor never gets tired of being warm. So the zine now says the resemblance fails there and says that is the more useful half. The same thesis supplied a better opening than the one already drafted: a spread that had begun with objections we invented now begins with what her interviewees were actually told. Both scheduled tasks also ran unattended — the weekly ingest at 9:01 and the system audit at 12:01 — the latter leaving two skill-edit candidates in a review file awaiting sign-off, which is the workflow behaving exactly as designed rather than a backlog.
Insights and Takeaways
- The one change we wrote down on Sunday fired on Monday, found two dozen faults, and immediately showed us the clause it was missing. Sweep for the number before closing the task turned up drifted counts on seven collection pages — a page saying “eleven collections” where there are thirteen, “forty-five zines and twelve field guides” where it is fifty-six and fourteen, a reading chain advertised at seventy pages that measures a hundred. It also caught one that was not stale but wrong from the day it shipped: a paragraph opening “No. 45 is the most lopsided chain here” about a zine that is not a chain and is not in that collection. All good. But the uncomfortable find is the one that revises the rule: a count corrected on Sunday was still wrong on Monday. The Sunday fix had been applied to the two pages that prompted it and never swept across the rest, so the corrected figure inherited the original defect and shipped with a fresher date on it. That is worse than leaving it alone, because a recent date is evidence that somebody looked. So the change needs a second clause it did not have: when a category turns out to be drawn too narrow, re-derive the whole set — patching the instances you noticed reproduces the bug. And a third thing the sweep exposed: a count is not one fact in one place. One figure was being printed in the meta description, the Open Graph and Twitter cards, the JSON-LD, the colophon and the body, so a search result and a link unfurl were both quoting a number the page had outgrown. Sweeping the visible prose would have missed five of the six copies.
- A gate can be working perfectly, reporting honestly, and still be measuring the wrong kind of number. The search index builder prints a coverage figure per page, and it printed 93% beside the collection page whose entire argument — the prose explaining why those pieces belong together — was not in the index at all. Nothing lied. The percentage was correct. The defect is that coverage is a ratio, while the thing the remainder record protects is an absolute quantity of prose: the larger a page’s repeating unit grows, the more text the same percentage conceals. So the safeguard leaked worst on the pages that scored best, and it leaked as a cliff rather than a slope — 93%, 97% and 100% on the three highest-scoring pages, hiding 8,438 characters between them, including every section intro on the front page. The page was penalised for succeeding: it only crossed the 90% threshold by growing to the largest card grid on the site, and two more pages were queued to lose their intros silently on the day they gained a member. This is a different animal from a green light is a claim, which we learned in Week 32 from checks that were broken. This check was not broken. It was the wrong shape of number for what it was guarding, and the only way to find that was for a person to ask what the index actually held rather than reading the column where 100% was ordinary. All six gates passed the whole time, and all six are structurally blind to it — a collision is not a colour, and an unindexed paragraph is neither.
- Twice this week our own working notes have been the source of an error rather than a guard against one, and that is the most expensive place for a wrong number to live. Writing Starlight — an essay whose subject is that working notes are a lead and not a fact — turned up a stale figure inside our own stylesheet comment, two numbers conflated from a fault we had corrected on the page but not in the code. Then the counting sweep found the repository’s own guidance file carrying the same drift: a collection listed at fifteen members, a guide count two behind, a reading chain six pages short. That file is what every future session reads first, which makes it the highest-leverage place in the whole system for a number to rot — and it had been rotting quietly while we built six gates to check everything else. It now carries a section saying plainly that every count in it is a lead and not a fact, with the command to re-derive each one and a note on which figures are historical and must be left alone. That is the standing rule we apply to aggregators, turned on ourselves. Which is also the week’s shape arriving a third time, from a third direction: the masthead we misquoted came from a sign-off in these notes, the love-locutions chain closed on our own page, and now the notes themselves. We are consistently our own least-audited source, because everything we write down starts out true and nothing schedules its re-checking.
Gratitude
- María del Carmen Rosas-Pérez, whose 2026 thesis arrived through our own inbox and then argued with the piece it was supposed to support. She put the question to 311 noise-sensitive people, and the finding that mattered — “no habituation, but accumulation effect” — is the one that breaks the analogy the chain was built on rather than confirming it. Her interviewees also replaced objections we had invented, including Participant 10’s sentence: “Well, it doesn’t bother me and I can’t see it, so it doesn’t exist and I don’t need to do anything about it.” And the you’ll get used to it everyone repeats turns out to rest on five participants, none of them noise-sensitive, listening to two pure tones for a fortnight — careful measurement generalised far past what it carries and handed back as a character note, which is that whole collection’s subject arriving inside its own endpoint.
- Essy Knopf, for crediting the origin he borrowed from — which plenty of retellings do not. Egg, and getting your egg cracked, is trans community language; he extended it to Autistic and ADHD experience in March 2025 and named where it came from in the same breath. Going after the coiner found nobody: an Urban Dictionary entry from 2016, a subreddit from 2017, and Susan Stryker unable to date her own first encounter with it. So No. 57 prints the origin as unknown rather than promoting the most citable name into a coiner — the exact failure we committed twice the week before. A word this good with no owner is not a gap in the record. It is what community language looks like from the inside.
Tuesday — 2026-08-18
Accomplishments
- Two zines and a field guide, and the second zine is the first one built with Helen Edgar rather than after her. The Stone You Keep (No. 58) grew out of her reading of the otter cam that has been running quietly behind Solidarity Sesh for months, and then went past the essay into the papers, where the argument turned out to be stronger than the anecdote — otters who never use a stone are not the ones who failed to work it out, and the ones who do show less tooth damage, so the support is legible in the skeleton. The Rest You Keep (No. 59) is the other half of the same Saturday room, eleven spreads whose structure is hers, built from Reclaiming Rest and Going Deeper — and it refuses something the essay it came from did not: it will not call meltdown or shutdown rest, because describing a shutdown that way turns an injury into a strategy and then hands the person the bill for having needed it. A Field Guide to Resting (No. 17) catalogues fifteen ways of doing it, sorted by what carries the rest rather than by how much of it there is.
- The front page had quietly become a hundred and sixteen phone screens, and nothing here measures a scroll. Measured before anything was touched: 94,607 pixels at phone width, 73% of it card descriptions, with the tallest single card running three and a half screens to describe one zine. It grew the way the counts do — a paragraph at a time, each one defensible on its own — which is the failure mode this system keeps rediscovering in a new medium. Every card now leads with the piece’s own tagline, then a short summary, then a disclosure holding the full description unchanged; nothing was deleted. Down 54% on the index and from 141,724 px to 104,805 across all thirteen collection pages, with the CSS deliberately not shared, because each collection’s accent tint is the thing telling you which one you are standing in. The thirty-four pages with no tagline of their own had to have one written from what the page already argues rather than invented for the index. And separately, the two skill fixes signed off from last week’s audit went in — a link checker that had been reporting resolvable citations as broken on every run, and a session sync that mis-rebased links quoted out of a skill’s own documentation.
Insights and Takeaways
- Three days, three instruments reporting health, and each one was broken in a different way — which is the actual finding, not any one of them. Sunday’s guard refused the correct call while silently permitting the wrong one. Monday’s search-coverage figure was the wrong shape of number: a ratio guarding an absolute quantity of prose, so it leaked worst on the pages that scored best and printed 93% beside a page whose entire argument was unindexed. Tuesday’s was worse and quieter — the front page went from 111 search records to 14, and the builder printed its tidy per-page line and carried on. These are not three instances of one bug. They are a refusal, a metric, and a silent success, and no amount of adding gates covers the next one, because each was a novel way of being wrong. The one thing that did fire on Tuesday is the shape worth keeping: a check comparing the new record count against the committed one — not a level, a delta. A level tells you what is, and can always be read as normal. A delta can be surprised. Every gate here except that one reports an absolute and therefore has no way to sound alarmed at a number it has never seen before. That is a design note, not a to-do list: the cheap wins are the checks that ask what changed, and they are also the only ones that would have caught two of these three.
- We shipped, on a live page, the exact failure we spent the same session correcting in somebody else’s citations. Tuesday’s fact-check pass fixed two Tricia Hersey quotations that had been attributed to the wrong book, both traceable to a secondary essay nobody had gone past. And in the same session, on a page already published, we had asserted that about seven in ten in-flight sleep bouts were asymmetric — a figure in neither source, taken from a search-result summary and attributed to a review that had not been opened. The thing that makes this worse than an ordinary slip is that the contradiction was already on the page. The colophon carried the correct qualitative claim while the body asserted a percentage, and we had both in front of us and shrugged. So the lesson is not check your sources, which we plainly know and were actively doing to someone else’s work that hour. It is narrower: a disagreement between two parts of your own document is a finding, and it is the cheapest one available — no lookup, no primary, no network, just noticing that the footnote and the sentence do not say the same thing. A summary that agrees with a plausible number is not a source, and the reason that one got through is that it agreed with something we already believed.
- Helen caught three things on Tuesday, and all three are about what a sentence does rather than what it says — which is precisely the category no gate here will ever reach. The first was a phrase in a refusals set: “an animal you find adorable is an animal you have stopped taking evidence from.” You take evidence about an animal and testimony from a person, so written that way the sentence put the otter on the stand as a witness for us — the exact move the same refusals set forbids four entries later. A borrowed courtroom register did the damage, and it took a reader who works in the language of testimony to hear it. The second was structural: a frigatebird sitting between a human claim and its conclusion, doing argumentative work that the same zine refuses five spreads later in as many words — with a hedge note underneath admitting the problem rather than removing it. And the third was an absence — the house convention names the superpower and inspiration framings as a pair, this zine had declined the superpower and left inspiration standing, on a piece whose opening spread is about being watched. None of these is a broken link, a bad colour, a collision or a stale count. They are a piece contradicting itself in a way that reads perfectly well, and the only instrument that detects that is somebody reading closely enough to be bothered by it. Monday’s essay said machines hold ground and people take it. Tuesday was the demonstration, and the person was not us.
Gratitude
- Helen Edgar, who read three faults out of two zines in one day and was right every time. One was a word — evidence from where it should have been testimony — and catching it needed somebody who works in that language rather than somebody checking references. One was a bird doing a job the zine had already forbidden. And one was a refusal that was missing rather than wrong, which is the hardest kind to see, on a piece whose opening spread is about being watched. She also supplied its replacement in her own structure, and the closing line is kept verbatim: “Nothing here is here to make you feel good about watching us cope.” She is not proofreading these. She is arguing with them, which is a different and much rarer offer.
- The researchers who published the range instead of the mean. Hall and Schaller counted 35.5 blows to open a mussel — and printed 6 to 88, so the range is the finding and the mean describes no mussel anyone ever opened. Harding and colleagues went back to the sleepiest-mammal statistic everybody quotes and found it was one measurement in a room at 33 °C, worth 11.0 hours at 26 °C, resting on a single study with an unknown number of specimens. The swift study reported that some individuals roosted for whole nights in midwinter while others never landed — which is what makes it a catalogue and not a ladder. And Tinker and colleagues followed one otter population closely enough to tell individuals apart and found it fanning out into specialists rather than converging on one best way to be an otter. Four studies, one habit: they refused to average, and each refusal is the reason the finding survived contact with us. The averaged otter forages nowhere.
Wednesday — 2026-08-19
Accomplishments
- Two zines and a field guide, and the first of them finally spends a naming we had been holding for weeks. Beyond the Optical Image (No. 60) takes Helen Edgar’s Dark Matter Field — the first of her twenty constellations — which Nos. 45 and 46 had each said in their colophons was being deliberately kept back. Its argument is about where a discrepancy is allowed to live: the mass, the law, or the measurement. Astronomy spent ninety-three years arguing all three out loud, and asked about a child the same question was closed on day one with the residual placed in the child. Thirteen Changes (No. 61) came out of tonight’s Infodumplings prompt and found, behind a one-line news story, a 298-page paper that published its calibration and then fenced its own instrument — and an appeal its authors filed and lost in print, with a number on it. A Field Guide to Unfinished Animals (No. 18) is its companion and adds one thing to the house form: every card carries a status, because a catalogue of corrections that will not say which corrections are finished reads as a list of settled triumphs.
- A publisher moved their writing and fifty-three of our citations quietly stopped resolving. Human Restoration Project retired a subdomain, and the obvious repair — keep the path, swap the host — would have 404’d on half of them, because three of seven articles had been renamed rather than merely relocated, including the one we cite twenty times. The fix came from pulling HRP’s own writing index rather than trusting a guessed address, with every destination fetched and its title matched against the link text we already had before anything was touched. Thirty-six occurrences now point at live pieces; two articles genuinely are retired and cite verified Wayback snapshots with their authors’ bylines intact; seventeen wait on hand edits because our publishing connector cannot reach glossary entries at all. Elsewhere the same day: the AI alignments page gained a twentieth entry — the first that maps a thinker we were already built on — and the Infodumplings page got two passes, the second of them cleanup rather than addition, which is the rarer and better kind.
Insights and Takeaways
- The premise that justifies making a piece is the one part of it nobody re-checks. No. 60 was proposed on a specific claim — that the field is “never directly observed, only detected by its distorting effect,” with gravitational lensing offered as the proof, and the parallel to neuronormativity called an exact structural match. Lensing is the instrument that made the direct detection. An eight-sigma spatial offset in the Bullet Cluster in 2006 closed the door Rubin had held open six years earlier, so the sentence that got the zine commissioned was wrong in the direction that made it worth commissioning. It is on spread seven now instead of in a private edit. What is worth carrying is not the physics but the position the error occupied. Every checking habit this project has runs on claims inside a piece — the quotation, the figure, the attribution, the joint between two rungs. A premise arrives before the checking starts, does the work of persuading somebody the piece should exist, and is then treated as settled by everyone downstream, including the person who wrote it. It is the one assertion in the whole process with no reviewer, and it is load-bearing for all the others. If a pitch is good enough to commission work, it is good enough to fact-check first.
- The fourth uncredited load-bearing term in nine days, and the mechanism is now legible: frequency is what hides it. Neuronormativity appears on fifteen pages of this project and had no attribution anywhere in it — it is Nick Walker’s neuronormative paradigm, and the source that settled it is Helen Edgar’s own writing rather than anything of ours. After Ze Frank, Discworld and the caves-campfires-watering-holes chain, the pattern has stopped looking like four separate oversights. A term used once looks borrowed. A term used fifteen times looks like vocabulary — it stops reading as somebody’s coinage and starts reading as the language the subject is simply discussed in, at which point nothing in a citation habit can be triggered by it, because our citation habit points at claims and a word is not a claim. The uncomfortable corollary is that the strength of the signal runs backwards: the terms most central to how we think are the ones we have said most often, which makes them exactly the ones least likely to still look like a debt. So the ones still outstanding are not the obscure ones. They are the ones that feel most like ours. The credit was also logged honestly rather than confidently — the book itself has not been read, so the colophon says confirmed by corroborating secondary accounts rather than implying we opened it.
- The machine found the breakage and every decision after that point was judgment — which is the cleanest statement of the split this week keeps circling. Fifty-three dead citations is a mechanical finding: a URL either resolves or it does not, and no reading is required. Everything downstream of it was a question no checker can hold. Was the article moved or renamed? Retired or relocated? Does the reader want a live link or a verified archive snapshot? Three of seven had been renamed, so the obvious repair would have failed on half of them, and the wrong reading of the evidence — a 404 means the article is gone — would have frozen twenty-five live citations into 2022 archive snapshots, burying work whose authors are still publishing it. A 404 is evidence about a URL, not about an article. Set that beside three days of instruments reporting health while broken — a coverage ratio guarding an absolute quantity, a record count dropping 87% without complaint, a nesting check that enumerates tag names — and the line between them is legible. The checks that worked this week tested something external and binary. The ones that failed tested something internal and continuous, where “normal” is a range the instrument itself defines and can therefore never be surprised by. That is not an argument for fewer gates. It is an argument for knowing which kind you have built, because only one of the two can be trusted to raise its hand.
Gratitude
- Vera Rubin, for withholding at the top of her own field. In 2000, as the most-cited person alive on the question, she was still writing that “we cannot yet rule out a modified gravitational potential, rather than dark matter, as the explanation of our observations” — and, about the object she had spent a career measuring, that “it is folly to believe that we know what a galaxy is, while the extent, the density distribution, and the composition of ≥90% of its mass are still a dark mystery.” Twenty years earlier, the paper the whole story hangs on had already refused the tidy version: not the curves are flat but they do not fall, not dark matter but mass beyond the optical image — a statement about where that declines to say what. Everyone downstream supplied the certainty. She never did, and it is her sentence the zine is named after. Zwicky did the same thing in 1933 and closed on “an unsolved problem” — two people forty-seven years apart, both holding a discrepancy open rather than placing it.
- Tschopp, Mateus and Benson, for publishing the calibration and then fencing their own instrument. Two hundred and ninety-eight open-access pages, and the methods section opens by conceding the soft part rather than hiding it — separation “is somewhat arbitrary… subjectivity of these interpretations is especially great.” Then they counted the differences between the pairs everybody already agreed were one genus, printed the numbers, set the bar one step past the largest agreed gap, and declared the result non-transferable in advance: it “cannot be applied to any other analysis, even of the same clade.” A threshold that names its own range of validity, inside the paper that depends on it. And when they wanted a second thing fixed they filed Case 3700, and the International Commission on Zoological Nomenclature refused them in print on the last day of 2018, in a numbered Opinion anybody can cite. Published provenance, a declared range, named dissent, a route of appeal, a citable ruling — five properties, and not one of them attaches to a label put on a child.
Thursday — 2026-08-20
Accomplishments
- Thirteen hours in Discord with community, and not one of them produced a file. A two-hour editorial meeting, an eight-hour career counseling session, and a three-hour Infodumplings whose cozy overflow ran two of them — the session that came out of Do You Still Have a Favorite Dinosaur?, covering a wide range of ground once the prompt had done its job. The overflow lasting longer than the thing it overflowed from is not a scheduling failure; it is what Impact means when it says an event booked for an hour often runs eleven, and it is the whole reason nobody has to be on time. A watering hole does not have an agenda, and the two hours after the topic ran out are the part that makes it one.
- We built a room for the credit that never gets given, and we stole it from WordPress and said so. A new #props-and-gratitude channel, where props is short for proper respect — borrowed straight from WordPress core, whose contributor attribution handbook is explicit that credit goes to design, testing, writing, user testing and well-written bug reports, and whose guiding rule is to “err on the side of giving props liberally.” The channel description points at that handbook rather than presenting the idea as ours, which is the practice applied to its own founding. What it is for is the work that does not ship: a bug report, a link, a proofread, a ride, a hard conversation, a day somebody covered for you. One line is a full props. No leaderboard, no points, no obligation to reciprocate, no
@required if you are not sure somebody wants to be named in public, and self-props welcome. The sentence from their handbook worth keeping is the one that reframes the whole thing: recognition is not a bonus for exceptional work. It is maintenance — the thing that keeps a project alive, because people who feel recognised keep showing up.
Insights and Takeaways
- The fullest day of the week produced the emptiest record, and every instrument was right about that. Yesterday’s only commit was the six o’clock notes backup — a timer firing, not a person finishing something. The gap-checker says plainly that commits are a floor rather than the day, and yesterday the floor was zero while the hours were the longest of the week. Which is the exact inverse of what the last five days kept teaching. Sunday through Wednesday we found instrument after instrument reporting health while broken: a guard that refused the right call, a coverage ratio guarding an absolute, a record count dropping 87% without complaint. Thursday every instrument reported nothing, correctly, and was still wrong about the day — because there was nothing of the kind it measures to find. So the failure is not calibration, it is shape: every check this project owns is anchored to an artifact, and a day made of attention leaves none. A record built from commits will always describe a Wednesday better than a Thursday. The honest response is not to instrument the gathering — that is the next paragraph’s problem, and the answer there is no. It is to notice that the reflection is the only part of this system that can hold a day like yesterday at all, which is a better argument for keeping it than any of the ones written down when it started.
- The thing Impact refuses to count spent eight hours doing it yesterday. Two weeks ago we rewrote that page to say we do not track the non-monetary mutual aid and are not going to start: it is organic, relational and continuous, there is no moment a check-in becomes an event and no line where finding somebody a clinic ends and staying in touch begins, so any number we produced would be a fact about our record-keeping rather than about the care. That was written as a principle. Yesterday it was a single session consuming a working day and then some — longer than the editorial meeting and the Infodumplings put together. The refusal still holds, and yesterday is what makes it worth restating rather than softening: an eight-hour figure would have described the clock and nothing that happened on it. But it is worth being clear-eyed that the refusal costs something, and the cost landed yesterday. The largest thing that happened is legible only to the people who were in it. It does not appear in a commit, a changelog bullet, a metric, or a funder brief — and the same page that declines to count it is the one that has to persuade somebody to fund it. We chose to be legibly uncountable. Yesterday is what choosing that actually feels like on a Thursday.
- The same week we kept finding uncredited sources, we built a room for uncredited labour — and it took writing them side by side to see they are one problem. Nine days produced four load-bearing terms with no attribution anywhere in the project: Ze Frank behind six zines, Discworld behind the method itself, Thornburg behind caves and campfires, and neuronormativity on fifteen pages. Every one of them had stopped looking borrowed. Then yesterday, with no reference to any of that, a channel for the credit that never gets given at all — a proofread, a link, a ride, a hard conversation, a day somebody covered for you. A citation and a props are the same sentence in two registers: this did not come from nowhere.* And they fail the same way. A term said fifteen times reads as vocabulary rather than as somebody’s coinage; care given continuously reads as the weather rather than as somebody’s labour.* Frequency and continuity are the two ways contribution disappears, and neither is anybody deciding to withhold credit — which is why both need a room built for them rather than good intentions.** The handbook line is the load-bearing one: recognition is not a bonus for exceptional work, it is maintenance. That is exactly what the fact-check ledger is for the sources, and we had not noticed we were building the second half of the same thing.
Gratitude
- The people who stayed for the overflow. The prompt did its job and then ran out, and two more hours happened anyway, across a wide range of things nobody had planned to talk about. Those two hours are unbookable — you cannot schedule the part where the agenda is finished and people are still there — and they are the part that makes it a watering hole rather than a meeting. Everyone who lingered was doing the actual work of the thing, and none of them will read it that way.
- WordPress core’s contributor attribution handbook, for writing down something most projects only gesture at. Its rule is to “err on the side of giving props liberally”, its list of who gets credited includes design, testing, writing, user testing and well-written bug reports, and its reasoning is not sentimental: contributors who feel recognised are more likely to keep contributing, so recognition is maintenance rather than reward. Twenty years of a volunteer project worked that out and then published the practice instead of keeping it as culture. We took it whole, and the channel says where it came from — which is the only way to borrow an idea about credit.
Friday — 2026-08-21
Accomplishments
- The first Campfire that asks people to move rather than watch. Seated Tai Chi, Transformative Somatic Practices, and Bodily Survival Knowledge went up Friday morning for Sunday’s session, and it is a practice, a reading, and an argument, in that order — which is a different shape from every Campfire before it. Twenty-three minutes of seated tai chi and qigong first. Then Nick Walker on transformative somatic practices, his dissertation already in our library, alongside a ninety-minute conversation between him and Katy Higgins Lee that puts somatics and neuroqueering in the same room. Then a punk happy flappy chair stim dance to Screaming Females covering “Shake It Off,” then a flamenco farruca to steal the hand movements and the foot stomps from, seated, and a close on a music video by our friend Scarlet Monk. Bluesky and Discord announcements went out with it. The phrase carrying the whole thing is bodily survival knowledge, and it is not decorative — the session’s argument is that this is knowledge, held in bodies, and that seated is not a lesser version of it.
- Friday’s other work was drafting a Campfire for somebody who died. Unlearning Toward Hope mourns and celebrates Professor Jason Arday with two of his own talks — one on neurodiversity, race and belonging, and one on what happens to a good word after an institution gets hold of it — with the transcripts already read into the knowledge system. It went out early Saturday. Ryan’s note setting it up does not soften why it exists: “We’re also going to discuss the anti-DEI, anti-neurodiversity backlash that contributed to Jason’s death. Our community is reeling from all the attacks and bad news.” The post’s own line is “What we are bereft of is care,” and it says plainly that we are doing this because he is gone, and because of how he was taken. The session also carries a definition worth keeping, credited to the Anglia Ruskin University Student Race Equality Advocates rather than absorbed: “Unlearning is the process of identifying an accepted reality that structurally disadvantages its inhabitants, some more than others.”
Insights and Takeaways
- A grief note became a library entry became a session, in eight days — and it is the first time all week the pipeline ran forwards. On Friday the 14th the note was four lines: “Pouring one out for Jason Arday. RIP,” an objection to the bigotry around it, and a Springer link. Checking it then, the garden held him once — a book chapter cited inside Helen Edgar’s writing — and stimpunks.org cited him nowhere. On Monday the 17th that paper went into the library inside a fourteen-file ingest. By Friday the 21st, two of his talks had been transcribed into the system and built into a Campfire. Eight days from a line written in anger to a session with the sources already on disk. Every other day this week the knowledge system was working backwards — finding a citation we had got wrong, a count that had gone stale, a term we had never credited. This is the direction it was actually built for, and the ingest preceded the work by four days rather than being assembled around it afterward. The measure of a library is not what it holds. It is whether the thing you need is already in it on the day you need it.
- Two Campfires drafted in one day, and their registers are opposite — which is not a scheduling accident, it is the week. One asks people to move: a seated tai chi class, flamenco footwork stolen sitting down, a punk chair stim dance. The other is mourning, and says on its face that it exists because of how he was taken. Same Sunday hour, two weeks running, and the note behind the second one says the quiet part: our community is reeling from all the attacks and bad news. It would be easy to read the practice session as the light relief and the memorial as the serious one. That gets it backwards. The post’s own phrase is bodily survival knowledge — survival, not wellness — and putting a twenty-three-minute somatic practice in front of a community absorbing that kind of news is the same argument the memorial makes, delivered as a body rather than as a talk. A movement class is what care looks like on a week when what we are bereft of is care.
- Two days running with one commit each, and the changelog holds both of them anyway — because a person wrote them in. Thursday was hours in Discord; Friday was two event posts, two social announcements, and a body of transcripts read closely enough to build a session out of. The only commit on either day was the six o’clock timer. An artifact-driven log would have shown nothing for either, and the skill that feeds the running file logs artifacts. What filled them was Ryan, by hand, on the day — so the day sections did exactly what they are for: the gap was visible, and being visible is the whole reason it got closed. That is Monday’s line about machines holding ground and people taking it, arriving at the record-keeping instead of at the pages. And a correction that belongs in the body rather than a footnote: this bullet first claimed both days were still missing and that Sunday’s draft would arrive two sevenths short. It was written from a check made an hour earlier and never re-run, while the entries were already in the file. On a day whose other insight is about an instrument reporting confidently from a stale reading, the reflection went and did it too. A level tells you what was true when you looked. Only looking again tells you what is.
Gratitude
- Professor Jason Arday, whose No one can see me cry sat outside this library until a week ago and is in it now — on mental health for Black and minority ethnic staff in higher education, which is a subject he was writing about from inside it. Sunday after next is built on two of his own talks rather than on anything said about him, which is the right way round. He should be here to be annoyed at how we summarised him.
- The people whose work makes Sunday’s hour possible, none of whom will be in the room. Nick Walker, whose dissertation on transformative somatic practices is in our library and whose conversation with Katy Higgins Lee puts somatics and neuroqueering in the same room. Screaming Females, for a cover loud enough to stim a chair dance to. Whoever dances the farruca we are going to steal the hands and the foot stomps from, seated. Scarlet Monk, closing it out. And the Anglia Ruskin University Student Race Equality Advocates, for a definition of unlearning good enough to build a session title on — “identifying an accepted reality that structurally disadvantages its inhabitants, some more than others” — credited to a student group by name, which is the week’s whole argument arriving one more time.
Saturday — 2026-08-22
Accomplishments
- A zine arrived as somebody else’s draft for the first time, and shipping it took six corrections and a restructure. The Light Arrived Before the Name (No. 62) was proposed, researched and drafted by Helen Edgar as a companion to her own writing on Grunya Sukhareva — the argument, the astronomers and the 1925 pairing are hers. The fact carrying it is that light leaves with nothing written on it, so a name in astronomy is an arrival record and never an origin record, and the piece is those two things being confused with a person in place of a star. Its spine is arithmetic: 168 years between one woman taking the Royal Astronomical Society’s Gold Medal and the next, 71 years for Sukhareva’s 1925 description to reach English, 123 years for the Matilda effect to be named — which makes the name for late naming itself a hundred and twenty-three years late. And the turn is that the DSM-5 is describing light-travel time when it concedes that symptoms may be masked by learned strategies in later life: somebody diagnosed at forty-seven has not been Autistic since they were forty-seven. The paperwork is forty-seven years old and the light is not.
- The Arday Campfire went out before dawn, and thirteen files got their index entries — including the sources Friday and Saturday had already been working from. Unlearning Toward Hope published at 4:39am, built from two of Jason Arday’s own talks rather than anything written about him. Both transcripts, and Nick Walker’s dissertation on transformative somatic practices, had been sitting in the library since Thursday lunchtime — which is why Friday’s Campfire post could say the dissertation was already in there. What Saturday added was the wiki entries: real titles, categories, and a place in the index for thirteen assets that the work had been drawing on directly out of the collection. The library had the files before the index had the names, which is the right order and worth noticing anyway, because it means the thing that lags is the finding rather than the holding. Also in that batch: the RE-STAR polygenic paper on adults who suspect they may be Autistic or have ADHD, a Lancet study on twenty-five years of recorded ADHD prevalence in England, both of the Yong Zhao essays that Wednesday’s AI alignments entry was built on, and a long-run study of what school suspensions do to adult outcomes.
Insights and Takeaways
- Three steps with Helen in eleven days, and the further the authorship moved to her, the more of the checking moved to us. No. 45 was written after her — her namings, our argument. No. 59 was built with her, eleven spreads whose structure was hers. No. 62 arrived as her draft, and shipping it took six corrections and a restructure. It would be easy to read that as the collaboration getting harder. It is the opposite. When we were writing after her, there was nothing of hers to correct — the errors available to us were our own, and we found plenty. Handing over a draft is a larger thing to offer than a finished piece, because it comes with the right to be edited attached, and the six corrections are what an editorial relationship looks like when both people would rather the piece were right than that nobody had to say anything. She has also spent the fortnight doing the same in the other direction: the evidence from phrase, the frigatebird carrying an argument it was forbidden to carry, a refusal that was missing rather than wrong. Four people can review a piece and only one of them will tell you the thing that costs something to hear.
- A zine about propagated miscrediting nearly propagated a miscrediting, and the fix was to leave the error on the page. The draft placed Sukhareva’s 1925 observations in Kiev, following a shorthand that travels widely in the secondary literature — she was born there and moved to Moscow in 1921, founding the hospital-school where the work was done. So the draft had caught the exact disease it was diagnosing: a claim that is repeated confidently enough to stop being checked. The reasoning for what happened next is the sharpest editorial call of the week. Quietly repairing it would have been the smoothing the zine exists to complain about — a piece about how credit goes astray, laundering its own strayed credit in a silent edit. So spread four now says what the draft said, where the error came from, and why it is still visible, the same way No. 1 keeps the sentence it got wrong. A correction printed beside its own cause is worth more than a correction, because the next person can see the trap rather than only the repair. Five more moved the same day, and the Annie Jump Cannon one is worth keeping because the corrected version is stronger than the draft: two honours had been merged into one she never received, and the true record includes Harvard withholding a regular appointment for forty-two years after she joined.
- The register test caught something no gate could, and it is the third kind of instrument this week — not a check, and not a person’s judgement, but a criterion written down well enough to hold a draft against. As No. 62 arrived, the light-travel fact opened and closed the zine and did nothing in between, which fails this collection’s own stated rule: one settled, checkable fact, followed honestly, already contains the claim about belonging. That is not a broken link or a stale count or a colour, so no gate could see it — and it is not a matter of taste either. It is a written-down structural requirement, and reading the draft against it produced a specific answer and a specific fix: the fact now carries every spread. Set that beside the rest of the week. The automated instruments failed in four different novel ways — a guard that refused the right call, a ratio guarding an absolute, a record count dropping 87% in silence, a nesting check that enumerates tag names. The catches that mattered were made by people reading, several of them Helen’s. Saturday shows the thing that closes the gap between those two, and it is neither more gates nor more vigilance. It is more criteria — rules explicit enough that a person applying one gets a structural verdict instead of an opinion. A gate tells you a page is wrong. A criterion tells you what shape it was supposed to be.
Gratitude
- Helen Edgar, for handing over a draft rather than a finished piece. A finished piece asks to be published; a draft asks to be edited, and the second is the larger thing to offer because it arrives with the right to be corrected attached. Six corrections and a restructure later, No. 62 is hers — the argument, the astronomers, the 1925 pairing — and she has spent the same fortnight doing the harder half in the other direction, telling us that evidence from was the wrong preposition, that a frigatebird was carrying an argument its own zine forbade, and that a refusal we had not written was missing. Three zines in eleven days went from after Helen to with Helen to by Helen, and none of that happens without somebody willing to be edited and willing to edit.
- Grunya Sukhareva, and the people who carried her forward. Six boys described in Moscow in 1925 — the aloneness, the flattened expression, the narrow blazing interests, and the giftedness underneath, in the same paragraph — published in Russian in 1925, German in 1926, and English in 1996, half a century after two other men’s names had been fixed to what she had already written down. She founded the hospital-school where she did it, and was attacked at the 1951 Pavlovian Sessions while colleagues around her lost their lives. And Margaret Rossiter, who in 1993 named the Matilda effect and thereby did the one thing the pattern never does by itself: she looked back down the beam and put the earlier name on it — Matilda Joslyn Gage’s, from 1870. The translators are the reason any of this is a fact rather than a rumour, and none of them are in the title.
Looking Back
Meaningful Moments
Publishing the piece on Jason Arday and the woes our community is facing. It felt good and proper to get that out there.
Life is Teaching Me
To fight. To fight for our community.
Habit Insights
Both the changelog and the reflection habit are proceeding apace.
Looking Forward
One Change I Can Make Next Week
Rest more. Do less.
I’m Looking Forward To
Relative rest during a rest week.