This log is backfilled from the collection's full commit history and kept up from here. It records three kinds of change: pieces added, pieces substantially revised, and fact-check and attribution audits — including the errors we found in our own work and exactly how we fixed them. Small typo passes and styling tweaks are left out; anything that changes what a piece claims is in.
We braid real science with ideas credited to named thinkers, so the facts and the attributions have to be right. We're human, and we've gotten things wrong — a paraphrase dressed as a Sagan quote, a Martin Luther King Jr. line credited to Baldwin, a forest-wide fungal network asserted as settled fact because it rhymed so well with mutual aid. Naming those in public is part of the method, not an embarrassment to bury.
The working guidelines and the full per-piece ledger live in FACTCHECK.md. If you spot an error, open an issue or reach us at stimpunks.org. Corrections are mutual aid.
Glimmer Wire, edition eleven: six papers read in full, and three of them held
The eleventh edition of the weekly graded scan is live — eight items filed, six held, three seeds, lags from 4 to 232 days with a median of 29.5. Six versions of record were read end to end, the most in eleven editions, and three of the six are held items rather than filed ones. Holding is not what happens when we cannot get the paper; this week it is mostly what happened when we could.
Graded verified, all three read end to end at PubMed Central: Boyette et al. in Molecular Ecology, who swab 327 museum frogs collected across the Brazilian Atlantic Forest’s unexplained 1970s decline and find “no evidence of Bd infection prior to the decline, followed by widespread infection immediately thereafter” — nothing before 1979, 58.8 per cent in the 1980s — while stating in their own conclusion that “by focusing exclusively on frogs collected at Boraçeia, we cannot extrapolate our findings to the broader Atlantic Forest”; Foo et al. in Ecology and Evolution, who track individual crown-of-thorns feeding scars at low seastar density and find that “26% of scars showed signs of regenerating coral tissue” by 22 weeks, a result the wire converted from a hedge about community diversity into a verdict that the reef’s most notorious animal is “actually helping it”; and Fu et al. in Nature, who recover enamel proteins from six Middle Pleistocene Homo erectus specimens in China and argue that the super-archaic stretches of the Denisovan genome, “some of which later passed to modern humans, are likely to have originated from H. erectus”.
Three are plausible on complete deposited abstracts, because none of those publishers would serve the paper. Zhang et al. in Molecular Cell describe type VI retrons, which “do not produce detectable msDNA in the absence of phage yet still provide defense” — the molecule used to detect such systems being absent, the lead author is quoted in the Gladstone release saying “we wondered if they might be broken” — and find the DNA derepresses a toxin where every other retron’s restrains one. Caulus et al. in New Phytologist rotate poplar stems inside a uniformly lit sphere to cancel gravity and light, and watch them counter-curve anyway by laying tension wood “on the convex side of curved stems until straightness is recovered”, which they say revises “our textbook view of tension wood formation”. Koga et al. in Nature Astronomy report “all five canonical nucleobases” in Hayabusa2’s Ryugu samples, concluding only that this “reinforces the hypothesis” of asteroid contribution to Earth’s prebiotic inventory.
Two are contested, and both turn on one word. Jost et al. in Science report a chiral magnon in “the altermagnetic candidate manganese telluride”, which the wire called conclusive proof of a third type of magnetism. Grice et al. in PNAS report heme-derived tetrapyrroles in a 115-million-year-old theropod bone, in a paper whose own first sentence is that “detection of endogenous biomolecules in vertebrate fossils remains controversial” — a live dispute the coverage does not mention, under the headline “dinosaur blood”.
The two we took most care with are about us and about children. A Science Advances paper titled “Maternal-fetal immune conflict contributes to male-specific impairments in a mouse model of neurodevelopmental disorders” was carried as a clue to autism’s origins; what it measures is fetal weight and sensory organ development in mouse embryos 24 hours after an immune challenge, and its own closing phrase is “acute, sex-restricted developmental deficits”. We read every word of it and are not grading our own existence as something a body is derailed into — nor repairing it into a gift, which keeps the same ranking. And a JAMA Network Open cohort reporting associations between ultra-processed food intake, striatal response and adiposity gain was headlined as a “troubling brain response”; the word is the headline’s alone, and it relocates a property of a food system inside a child.
The sharpest contradiction this page has found between a release and its paper. A Texas A&M release says a “major new study challenges these assumptions” about reading; the paper’s own abstract says “the findings were generally consistent with previous research”. It is a cross-sectional meta-structural equation model over correlations among test subtests, it followed no child and ran no intervention, and it was published on 16 August 2025 — a lag of 444 days, the longest this scan has recorded by journal date.
Also held: a PNAS paper on foveolar cone patterning, read in full, in which the string vitamin A occurs zero times and retinoic acid eleven, carried as a vitamin A discovery 260 days late and conducted in organoids; a Physical Review D survey of the conditions under which black holes could fragment — all of which require higher dimensions or modified gravity, the abstract’s second sentence being that the area law “forbids such processes” in classical general relativity — reported as a loophole somebody found; and a new genus of seahorse-lookalike fishes known only from museum drawers, held with nothing wrong with it that we could find: it is published in a journal whose site does not resolve from this container, so its editorial process could not be established, and soundness in taxonomy is partly a property of the venue.
Last week PubMed Central challenged every request; this week it challenged four of seven. We did not attempt to defeat the challenge and did not disguise the caller, which is the same rule as last week. The edition says plainly why the intermittency is the worse condition: a total block is a fact about the archive that reproduces, while an intermittent one means which three papers were read in full is partly a coin toss, and a reader repeating the work will get a different three.
And a mechanism Too Good to Check has no row for. The altermagnon result sat on arXiv from 29 January 2025 — 644 days before the wire carried it, against a journal lag of 6 — and the two abstracts differ: the preprint claims “the first direct experimental evidence” of “a new class of magnetic excitations”, where Science published “direct experimental evidence” of “a discrete class”. Peer review removed a priority claim and softened a novelty claim; the coverage then asserted something stronger than either. That is the opposite direction from the marmoset case recorded on 23 October, and the pair is filed as a seed rather than a finding — two instances is not a pattern, and the edition says so.
And one ordering fault in our own tooling, hit for the first time this week. llms.txt states how many of the site’s pages have a Markdown sibling, so it has to be written after build-markdown.mjs creates them — and the order this project has had written down since the siblings shipped runs build-derived first. It has never mattered before, because that count only moves when a page joins the Markdown set, which this edition did. check-derived.mjs --write caught it by checking its own work: it wrote llms.txt saying “70 of the 220 pages”, then generated the new sibling, then reported llms.txt STALE against the true 71 of 221. A second run of the generator fixes it, and the documented order should be amended rather than the habit of running it twice.
The picture census: eight ScienceDaily feeds carried content, deduplicating to 66 unique releases in the seven days to 6 November. Seven lead with an image credited AI/ScienceDaily.com and 33 more with a Shutterstock photograph — 40 of 66. Last week was 57 of 88 and the two must not be subtracted: the denominator moved by a fifth, and a ratio over a different mix is a different measurement.
The code goes MIT, and the collection stays CC BY-SA
From today the code in this collection’s repository is under the MIT License: the build and check tools, and the two scripts the pages share. Every zine, field guide, broadside, diagram and word stays CC BY-SA 4.0, exactly as before.
Why. Creative Commons advises against its licences for software, and the host this site runs on asks open-source projects for a software licence. It was Ryan Boren’s call, made across every Stimpunks repository on the same day, and Helen Edgar agreed to it for the code she has written here.
What is code: everything in tools/, and starstuff.js and edit.js. What is not, even inside those files: the words and pictures a script carries — the chain spines, the L★S mark, the prose the build tools write into derived pages — the scripts written inside a page, the stylesheets, the guest pieces under their own notices, the typefaces and anybody else’s pictures. The licence file now leads with the MIT section and keeps the full CC BY-SA 4.0 text, word for word, beneath it. Copies of the code taken before today may still be used under CC BY-SA.
One sentence of ours had to change with it. The design system page said it was CC BY-SA “like everything else here”, which stopped being true this morning; it now says the words and pictures are CC BY-SA and the tools that measure them are MIT.