The editions
Newest first. Each card gives the tally and the median lag; the Details hold what the edition got wrong and what it could not open.
Stimpunks Foundation × More Realms · Working Papers
Glimmer Wire
A Star Stuff collection · 8 editions · weekly, Fridays · started 2 September 2026
A weekly read of what is going well in science and medicine, graded before it is enjoyed. This is the scan that sits behind Glimmers, published with its grades showing rather than kept in a drafts folder. One page per edition, newest first, and nothing is ever deleted — including the thin weeks.
Our weekly reading, published as we do it. Each edition lists the science stories we looked at, says how far we actually got with each one, and prints the items we decided not to run beside the ones we believed.
Most of it never becomes anything else. That is the honest part.
Nothing here is a source; it is a list of leads. What the grades mean, and why each week gets its own page, are below the cards.
16 October 2026
Edition Eight
Glimmer Wire · 8 filed, 8 held
Nine papers read end to end, the most in eight editions — and a headline that puts a word in quotation marks which nobody on the page is ever quoted as saying.
A dinosaur that waited a hundred and two years in a museum before anyone established what it was, an octopus genome whose authors coined their own word for interdependence and then named what it costs, a mouse embryo that wrote down its own family tree as it went, and an assumption underpinning how we read every planet’s chemistry that had never once been tested.
Details
The lag ran from 0 days to 236, median 19, and the zero is real: Invictibellatrix amberina was described and reported on the same day, 14 October, after sitting in the Natural History Museum since 1924 — and the paper’s own methodological claim is that the braincase everyone had written off as phylogenetically conservative was the part that answered. Five verified, one contested, two plausible. The contested one is a hippocampus under propofol still following speech and getting better at it over ten minutes: the recording is solid, and whether it is a claim about consciousness is what the paper’s own first sentence calls disputed. Eight held. One headline says brain where its paper’s title says pharynx, with a quoted researcher as the only bridge. One scare-quotes the word astonishing, which appears nowhere on its page but in the headline and nowhere in the paper. One carries a finding from August 2023 in the present tense, lag 1,141 days — the third edition running to hold an item of that shape. One turns fourteen weeks of limb loading in growing fowl into advice about human childhood. Two are declined on our own grounds rather than the work’s: two cats framed as a disorder, and a measurement of genocide that is not a Glimmer and is not ours to summarise. Three of the eight held items were read end to end, which is the point of the section: holding is not what happens when the paper is out of reach, it is what happens when it is not. And one correction we nearly published ourselves: following redirects made doi.org look dead when the resolver answered correctly every time and the publishers behind it did not.
9 October 2026
Edition Seven
Glimmer Wire · 6 filed, 6 held
Five papers read end to end, and every one of them from the same archive — because four of their publishers could not be reached at all.
A deep-sea coral that three genes placed in three different families while each reported near-maximal confidence, a date for the origin of animals pinned to a deposit where nobody found any, a solution Hans Bethe wrote down in 1931 assembled out of caesium atoms, and a hundred-species survey whose own conclusion refuses both halves of the question it was asked.
Details
Every full text this week came through one host, and the reason is the interesting half. Five versions of record were read end to end, all five at PubMed Central — and four of the five are openly licensed at publishers (Pensoft, PeerJ, Oxford, Nature) whose own sites refused the connection outright. An open licence is a permission; this week it was not a road. Six items run 1 to 216 days, median 68.5, and for the first time in four editions the lags are spread rather than clumped at the ends. Four verified, one contested, one plausible. The contested one is the metazoan calibration: the argument that a maximum age built on an absence proves nothing is well made, but the 800-to-700-million-year date it supports is a clock estimate specialists will argue over for a decade, and the headline’s “200 million years” is in no abstract. Six held, four of them because the headline is not the paper — a magnetofossil whose imaged specimen comes from the Palaeocene–Eocene Thermal Maximum rather than the 97 million years in the title, and which the paper attributes to unidentified eukaryotes rather than an animal; a genetic-code discovery published in October 2023 and carried in the present tense, lag 1,098 days; a synthetic ruthenium complex reported as photosynthesis; and a Homo erectus date from 2020 reported as this week’s. And one measured finding of our own: of the 76 science-news items scanned, 14 lead with an image credited to a machine and 37 more with a stock photograph. The credits are honest; the ratio is the point.
2 October 2026
Edition Six
Glimmer Wire · 8 filed, 6 held
Six papers read end to end — and the first story held because the only summary of it available was written by a machine.
A shark that hears far past the limit inferred from an organ it does not have, eleventh-century children buried beside adults who were almost never their kin, a rare syndrome whose old prevalence figure turns out to have measured insurance criteria, and a scroll built and carbonised on purpose so that an algorithm could be marked against a known answer.
Details
The best reading week this page has had, and the reason is a repository answering on the second knock rather than any new skill of ours. Six versions of record read in full — four at PubMed Central, which served Google’s reCAPTCHA at HTTP 200 first and the papers on a retry, four times out of four, and two at PLOS. Eight items run 4 to 91 days, median 20, less than a quarter of last week’s, with four of the eight inside ten days. Five verified, one contested, two plausible. The contested one is the new bat phylogeny, where a European origin inferred from a fossil record that is best sampled in Europe is the kind of claim specialists argue about for a decade — and where the headline’s “65 million years ago” appears in neither the abstract nor the release, both of which say late Palaeocene, about six million years later. Six held, and two of them for numbers rather than words: a cannabis meta-analysis whose “twice as likely” is its cross-sectional figure while its longitudinal estimate is OR 1.17, and a brain-ageing study whose headline drops the result we would have led with, that autism and ADHD show no difference at all. A new kind of refusal: where a publisher elides an abstract, an index returns a fluent, paragraph-shaped, correctly labelled summary written by a language model in the same field an abstract would occupy. It is not a source, nothing about its shape says so, and the Nature paper it belonged to is held rather than filed.
25 September 2026
Edition Five
Glimmer Wire · 7 filed, 6 held
Five papers read end to end — and six holds whose headlines each turn on a word the paper does not use.
A result that finds disorder stabilising a network rather than wrecking it and traces the opposite belief to a shortcut taken for the maths, one genus of tree with three different ways of breathing, a penguin described this year from a skin collected in 1897, and the oldest solids in the solar system still holding a magnetic field stronger than Earth’s.
Details
The grade key stopped being a ladder this week. Four versions of record were read in full at PubMed Central — the PNAS paleomagnetism paper on calcium-aluminium-rich inclusions, the Nature Communications NUDT5 paper, the Nature Communications Clusia genomes, and the Communications Biology gentoo penguin paper — and the penguin paper is graded contested anyway, because it reports its own field as still disagreeing and reaches its answer partly by demoting a species another team described. A fifth paper, the Science result on disorder-promoted stability, was read in full as the authors’ arXiv preprint and graded plausible, which is the preprint rule costing this page a grade for the second week running. Seven items run 2 to 157 days, median 85. Six held, five of them for a headline word the paper does not use: a Stonehenge story whose abstract ends by saying glaciers alone cannot account for the stone, a whale paper titled post-mortem attendance reported as mourning, two progenitors reported as two organs, 83 hits from a scan of 1,049 outcomes reported as 83 benefits — and a new human blood group that is a paper from 2024, between 631 and 760 days old.
18 September 2026
Edition Four
Glimmer Wire · 8 filed, 6 held
The week four papers opened end to end — and the longest lag the axis has recorded, 330 days.
A single tile shape that orders a plane without ever repeating, an Earth that matches no meteorite we have collected, a textbook arrow that was assumed before it was calculated, and five crocodiles that each answered the question of home differently.
Details
The best reading week this page has had, and the median lag tripled because of it. Four papers read end to end, all four in PubMed Central under a Creative Commons licence: the monotile diffraction experiment in Nature Communications, the Earth-accretion reanalysis in Nature Astronomy, the inductive-effect paper in the Journal of Chemical Education, and a superficial-white-matter study in Alzheimer’s & Dementia drawn from a community cohort in India. Three more rest on publisher-deposited abstracts at Crossref. Eight items run 1 to 330 days, median 88.5, against 14 last week — the jump is this scan following citations rather than releases, and the 330-day item was public as a preprint 397 days ago. PubMed Central served a reCAPTCHA at 200 on about one request in three, so a run that trusted the first response would have reported four unreadable papers and exited clean. Six held, two of them papers we wanted: a “hundred-year” textbook correction whose evidence is 652 and 295 days old, four semaglutide stories in seven days whose loudest is female mice, a Venus moon the paper’s own title calls hypothetical, and a physicists’ survey held only because it exists as a preprint with no version of record.
11 September 2026
Edition Three
Glimmer Wire · 7 filed, 5 held
The week the network answered — three papers read end to end, and the longest lag the axis has measured.
Nine Jurassic species each singing at its own frequency, one gut bacterium that is really several populations, a cat that was never a cheetah, and two papers in one journal volume that disagree about whether the universe is accelerating.
Details
The first edition where the refusals mostly lifted. Three papers read in full, all three in PubMed Central under CC BY; a fourth read as the authors’ own preprint and deliberately not graded verified, because a preprint is the authors’ text and not the version of record. Still refused, and named: Cell, PNAS, Science, Nature.com, IOP, OUP, Europe PMC, the PubMed E-utilities and doi.org — plus PubMed’s own site, which answered every request with 200 and served a cookie notice instead of the record. The lag axis had its most useful week: 0 to 187 days, median 14, against 36 and 40.5 before; two items were public as preprints far earlier, one by 834 days. Five held, led by the more than 500 vesper bat species figure this page corrected last week and which is still on the wire, and by a 2022 mouse paper carried as this week’s discovery with the wrong institution credited.
4 September 2026
Edition Two
Glimmer Wire · 6 filed, 4 held
The week nothing could be opened, published saying so — and then amended when the block lifted.
Every publisher, preprint server, aggregator and DOI resolver was refused at the network layer before the request reached the site. Filed on search summaries, graded accordingly, and re-run hours later against the primaries.
Details
What the amendment bought was not confirmation. It was a list of things we had wrong. One item upgraded — the bats paper turned out to be open access in PubMed Central and was read in full — and corrections landed in all six entries. The two sharpest were ours, not the coverage’s. We wrote that a Science paper’s title said the impairment limits organ ageing and convicted a science desk of overreach on that basis; the title’s subject is restored clearance, so it says the opposite of our reading. And we opened a Hall-effect entry by describing the ordinary Hall effect (applied perpendicular field, Hall 1879, hence “century-old”) before reporting the paper as contradicting it — it is the anomalous Hall effect, and we reproduced a press release’s conflation in the act of complaining about it. Three more were attribution errors: a single-author review called “a NASA-funded group”; three dates printed as cloud encounters that are actually cooling intervals in deep-sea sediment cores; and a “more than 500 species” figure that appears nowhere in the paper it was credited to, where the measured number is twenty-six. Five of six are still not read at source, and each entry names the refusal: PNAS, Cell and Annual Reviews returned 403, Nature Materials served an abstract and priced the rest at $39.95, Science paywalls past its abstract. Getting past the network was not getting past a paywall. Four items held, including the one most on-frame for this whole site.
2 September 2026
Edition One
Glimmer Wire · 7 filed, 3 held
Seven items filed on institutional press releases, because six publishers refused a direct read that day.
Saturn’s ten-sided south pole against its famous hexagon, rocky planets a hundred million years after the Big Bang, diamond turning pressure into charge at its grain boundaries, and the triune brain retired across 182 species.
Details
The edition that proved the ratio. Two of its seven filed items went on to become pages, and neither did so on the strength of what was filed here. The Saturn decagon was upgraded to verified the same day, when the paper was obtained and read in full — which corrected two things the NASA release had left ambiguous, including a “nobody knows why” that overstated authors who name two candidate mechanisms and say they cannot choose between them. It became Until There Were Ten. The triune-brain paper was read in full and became No. 85 — and corrected this digest’s own entry, which had the inverse relation running between the limbic system and the neocortex when it is specifically the olfactory system, with limbic components on the olfactory side. The fibromyalgia item stays contested on better grounds: both papers have since been read in full, and they disagree in print about sex, autoimmunity and heritability — so it is not unchecked, it is genuinely unsettled, and the piece is parked. Three held, and the first is the reason the section exists: a hundred-whale humpback aggregation, contradicted by the peer-reviewed survey of that same water, which recorded densities of 0 to about 0.12 individuals per square kilometre across fourteen austral summers.
How this collection works
What holds these together
Every other collection here holds finished arguments. This one holds the working stage of one: a scan of the week’s hopeful science, with each item graded, each grade justified, and each entry stating what was actually read at source and what was not. Most of what gets filed here never becomes a page. Two items have — Until There Were Ten and No. 85, both from the first edition, both only after their papers were finally obtained and read. That ratio is the point of showing the scan at all.
Items in an edition are sorted by lag — how long a finding had been public before a news wire carried it here. That number turned out to be the argument for a weekly cadence rather than a daily one: the medians so far are 36, 40.5 and 14 days, so a daily digest would mostly re-encounter what it had already seen. The third edition’s median is much the lowest, and the reason is not that the world sped up — it is that the scan finally reached journal feeds instead of reaching only what a search summary had digested. It is also an honest number to publish about yourself, and it has four times been found to understate the gap, because a journal date cannot see a preprint that was public earlier — by 288, 478, 753 and, in the third edition, 834 days.
The axis is issue, and that is a fifth one here
The five register collections sort by the kind of argument a piece makes. Field Guides and How We Got Here sort by form. Print and Sound sort by medium. Glimmers sorts by occasion — something happened on a particular day. This one sorts by issue: a member is here because it is edition N of a serial, and its identity is the date it went out.
So editions are dated, never numbered, for the same reason Glimmers entries are: the zine numbering says when and is referenced by number across the changelog, the fact-check ledger and years of prose with no redirects. A serial that inserted itself into that sequence would make the numbers mean two things. An edition’s address is its date, and its date does not move.
Why each edition is its own page
It was one page until 4 September 2026, with editions stacked newest-first behind a single masthead. Two editions in, that page was already long enough that the second one buried the first, and the shape gets worse with every week rather than better — a running log on one page is a page that grows without bound and re-reads as a wall.
Splitting it costs something real and worth naming: an edition no longer sits in the immediate company of the edition before it, and this scan makes cross-edition observations on purpose — the same held item declined twice, a lag distribution confirmed by a second data point. Those now have to be links rather than a scroll. In exchange the reader gets an edition they can finish, a URL they can send that means that week, and an archive that stays navigable at fifty editions instead of collapsing under them. The old single-page address, glimmer-wire.html, permanently redirects here.
What the grades mean, and why Held is the section that matters
Four grades, and they are about our evidence rather than about the science. Verified means the primary source was read and confirmed, and the entry says so in as many words. Plausible means the finding is sound and the metadata confirmed, but the paper’s text was not read — and the entry names the exact refusal that stopped it, whether a paywall, a 403, an authentication redirect or a reCAPTCHA. Contested means specialists disagree, or the claim outruns the method. Held means it could not be traced to a source that supports it, and it stays on the page anyway.
That last one is why the page exists. A fact-check ledger records what we checked; a changelog records what we got wrong and fixed. Neither records the story we decided not to tell you, and that is usually the most interesting thing about a week — the first edition declines a hundred-whale humpback aggregation that the peer-reviewed survey of the same water contradicts, the second declines the item most on-frame for this entire site because its sourcing traces to a congress abstract, the third holds a figure this page itself corrected the week before, still circulating unchanged on the wire we scan, and the fourth holds two papers it wanted to file — one because its “hundred-year” correction rests on evidence 652 and 295 days old, the other because it exists only as a preprint.
The standing refusal: this is not a citable source
Nothing filed here may be promoted to a zine, a Glimmers entry or a card without the primary source first, a changelog entry and its row in the fact-check ledger. The digest is a set of leads. Lifting a figure out of it and printing it elsewhere would be laundering a search summary into a fact, which is the precise failure this collection was built to make visible — and which the second edition committed, then published about itself.
Two of the eight editions have been amended after publication, and the amendments are on the pages rather than in a footnote. The second edition’s is the instructive one: re-run against the primaries, it upgraded one item, corrected all six, and the two sharpest corrections turned out to be ours rather than the coverage’s — we had misread a paper’s title, and described a physics result using the wrong physics. Too Good to Check names the ways a fact goes wrong here; this collection is where you can watch it happen.
Off the reading chain, on purpose
Nearly every page on this site is linked into a single prev/next reading order, from Start Here through to the last broadside. This collection and every edition sit outside it, for the same reason every working paper does: threading a weekly scan between two zines would drop a maintenance document into the middle of somebody’s reading. The editions carry prev/next to each other instead, which is a local order through the archive rather than a link in the site’s chain.