Stimpunks Foundation × More Realms · Working Paper

Edition of 11 September 2026

The third edition of Glimmer Wire, the scan that sits behind Glimmers — and the first one where the network answered. Three papers were read in full, all three in PubMed Central, all three the version of record under a CC BY licence; a fourth was read as the authors’ own preprint and is graded accordingly. Items are sorted by lag, which this week runs from 0 to 187 days, and the section that matters most is Held, which lists what we declined and why. One held item is a figure we corrected last week, still travelling.


7 filed · 5 held · 2 seeds · 3 verified · median lag 14 d

What was actually read: three papers in full, one preprint, three abstracts

Last week’s edition was compiled with every source refused at the network layer. This week most of the refusals lifted, and the tally moved with them. Answered when tried, by name: api.crossref.org, api.semanticscholar.org, pmc.ncbi.nlm.nih.gov, arxiv.org, sciencedaily.com, sciencenews.org and journals.plos.org. Still refused, also by name: doi.org, eutils.ncbi.nlm.nih.gov, www.ebi.ac.uk (Europe PMC), export.arxiv.org, api.openalex.org, phys.org, eurekalert.org, science.org, cell.com, sciencedirect.com, pnas.org, nature.com, iopscience.iop.org, academic.oup.com and news.ucsc.edu. One refusal is worth naming separately because it looks like success: pubmed.ncbi.nlm.nih.gov answered every request with HTTP 200 and served a cookie notice instead of the record, so a status code is not a source.

What that bought is three verified items, and it bought them in one place. PubMed Central holds author-accepted and publisher deposits of open-access papers, and three of this week’s seven are there in full and under CC BY: the Nature gut-microbiome paper, the PNAS Jurassic soundscape, and the Nature Communications eight-letter transcription paper. Those three are graded verified and the entries say what was read. A fourth, the interstellar-comet result, was read in full as the authors’ own arXiv preprint and is not graded verified, because a preprint read in full is the authors’ text and not the version of record, and this page does not blur the two. The remaining three stand on publisher-deposited abstracts.

One of those abstracts is a hop further out than usual, and it is flagged rather than smoothed. The American-cheetah paper’s abstract was obtained from the Semantic Scholar Graph API, not from Crossref and not from the publisher — Crossref carries the paper’s metadata but deposited no abstract, and Cell and ScienceDirect both refused the article page. Semantic Scholar redistributes a publisher deposit rather than authoring it, but it is one more link in the chain than we would like, and the entry says so.

The lag axis had its most useful week so far. The seven items run from 0 days to 187, with a median of 14 — against 36 and 40.5 in the two editions before this one. That drop is not the world getting faster; it is this scan finally reaching journal feeds rather than reaching only what a search summary had already digested. The longest item is the useful one: an Astrophysical Journal Letters paper published on 6 March 2026 and carried to us on 9 September as news, when the same result had been covered at publication six months earlier. And the axis is again structurally blind in the same way: two of the seven were public as preprints long before their journal dates — the gut-microbiome result by 834 days, the comet result by 288. Both entries print that interval as well.

A note on the wire dates, because they are easy to get wrong by a day. ScienceDaily’s category index lists three of these items under a date one day later than the timestamp in the release’s own URL. The lags below are computed against the URL timestamp, which is when the release was actually posted, and every entry prints the wire date it used.

Filed

Sorted by lag, shortest first. Bars are scaled to the longest item, 187 days; the three items at 0–2 days would draw a bar too thin to see, so they are given a visible floor of 2% rather than a true-to-scale sliver, which is a drawing decision and not a measurement. Four of the seven carry a date inside the last ten days. The other three are 88, 123 and 187 days old, and each was carried to us this week by a feed treating it as new.

Verified Primary source read and confirmed.
Plausible Finding sound and metadata confirmed; paper text not read.
Contested Specialists disagree, or the claim outruns the method.
Held Could not be traced to a source that supports it.

Plausible Published online 2026-09-04 Journal Current Biology Read the abstract, at Semantic Scholar

Lag 0 d

The American cheetah was never a cheetah, and it was not one thing either

Miracinonyx trumani has been read for decades as North America’s answer to the African cheetah, on the strength of a long-legged, cursorial build. This group generated nuclear palaeogenomes from specimens in Wyoming (~23,000 years old) and the Yukon (~31,000), and the phylogeny puts the animal sister to pumas, not to cheetahs, with a divergence around 2.6 million years ago. The Yukon material extends the known range by more than 20° of latitude into the Arctic. And stable isotopes separate the two populations sharply: the Yukon animals sit at an elevated trophic position consistent with eating anadromous fish, while the Wyoming animals were generalist terrestrial predators. The abstract also reports loss-of-function mutations in circadian genes, which the authors read as relaxed constraints on circadian regulation rather than as an adaptation. Its closing sentence is the one to keep: these findings underscore how morphological convergence can obscure the true diversity and complexity of extinct megafaunal ecosystems.

The star stuff frame Two things happen here and the second is the better one. A resemblance got mistaken for a relationship, and the name recorded the mistake for a century. Then, once the name came off, the animal turned out not to be a single way of living at all — the same species eating salmon under an Arctic light cycle and hunting on temperate grassland, with the genome relaxing the clock in the population that no longer needed it. The category was doing the flattening. That is the shape this site keeps meeting from the human end: a label built on surface likeness, and a range of real ways of being that only becomes visible once the label stops answering for them.
Four specimens, and a date we could not settle The palaeogenomes come from four specimens, one from Natural Trap Cave in Wyoming and three from the Yukon — a small sample for a claim about two ecologically distinct populations, and the isotope result is a dietary inference from tissue chemistry rather than an observed meal. The salmon reading is the authors’ own and is hedged in the abstract as consistent with. The publication date is genuinely unresolved and the lag inherits that. Crossref gives the paper a print date of September 2026 with no online-first date, and registered the DOI on 4 September 2026 — the same day the coverage ran, hence the 0 days. But the same Crossref record carries a CC BY licence whose version-of-record start date is 31 July 2026, which would make the lag 35 days instead. We could not adjudicate, because the publisher’s own article page is refused. The 0 d is therefore a floor, not a measurement, and it is the smallest number in the sort for a reason that may not survive contact with the page we cannot open.
Where it goes Checked against the shelf rather than assumed: convergent evolution is not unspent here. No. 31, Five Times a Crab, Seven Times Not, is built on carcinisation, and Too Good to Check and the watching-animals guide both name it too. What is unspent is the second half — one species, two ways of making a living, separated by 20° of latitude — which is a different argument from convergence and a better fit for More Than Human than for a chain. A chain link is possible on the naming error alone and would be marked documented.

Cassatt-Johnstone, M., Wooldridge, T. B., Ghosh, S., Hall, E., Hewitson, S., Meachen, J., Wooller, M. J., Zazula, G. & Shapiro, B. “Range and diet diversity in the Pleistocene American ‘cheetah,’ Miracinonyx trumani.” Current Biology, 2026. DOI 10.1016/j.cub.2026.07.072 · PMID 42697199. Read: the paper’s abstract, as redistributed by the Semantic Scholar Graph API from a publisher deposit — one link further out than this page likes, and the reason this is not graded verified even though the paper is CC BY. Not read: the paper. Cell returned a 403, ScienceDirect was refused at the network layer, and doi.org returned a 403; there is no PubMed Central copy. Not the paper’s own: nothing in the finding above — but the specimen provenance (Natural Trap Cave; three Yukon specimens) comes from the coverage, not from the abstract, and is marked here rather than carried as if it were. Wire: Science News, 4 September 2026.

Verified Published online 2026-09-02 Journal Nature Communications Read the paper, in full

Lag 2 d

A cell’s own transcription machine reads four letters it has never met, the same way it reads the four it has

Every living thing on Earth writes with four DNA letters. The hachimoji system adds four synthetic ones in two extra pairs, P:Z and B:S. This group shows that Escherichia coli RNA polymerase — an ordinary bacterial enzyme, not a redesigned one — transcribes the full eight-letter alphabet efficiently. They hit one real failure, G:Z misincorporation, and fixed it by synthesising a higher-fidelity analogue, Z*, with the C5 nitro group replaced by a carboxamide. Then four cryo-electron-microscopy structures at 2.42–2.75 Å show what the enzyme is doing: in the paper’s words, E. coli RNA polymerase recognises these unnatural base pairs “in the same manner as natural base pairs.” Neither P:Z nor P:Z* incorporation impairs the elongation that follows. The A-locus arithmetic of the enzyme did not have to change to accommodate a doubled alphabet.

The star stuff frame The reading that matters is not we can build new letters. It is that the machinery was never as narrow as the alphabet it evolved on. A polymerase shaped by four billion years of exactly four characters turns out to accept eight without being asked twice, because what it checks is geometry rather than identity. A system that looks purpose-built for one arrangement was actually built for a shape, and the arrangement is what happened to arrive. That is the difference-first claim in its most literal form: four letters is a contingency, not a requirement, and the thing doing the reading knew that before we did.
This is a test tube, and the headline is not The ScienceDaily headline is Life uses 4 DNA letters. Scientists just made 8 work. Nothing here happens in a living cell. The transcription assays are in vitro, on synthetic nucleic-acid scaffolds, with E. coli RNA polymerase expressed and purified from plasmids — the paper’s own closing claim is that the results “establish the feasibility of an eight-letter genetic alphabet for transcription,” which is a sentence about feasibility and about transcription, and about neither life nor cells. Nor is the eight-letter alphabet new. Hachimoji is Steven Benner’s system, published in 2019, and Benner is an author here; what is new is that a multi-subunit cellular RNA polymerase handles it, the structural account of how, and the Z* analogue. We have not read the 2019 paper, so its date is reported from this paper’s own citation of it and not independently confirmed. One more thing the frame above must not be allowed to carry: an enzyme accepting a synthetic base pair says nothing about whether any organism could live on one.
Where it goes Unspent, and by a wide margin: no page here contains base pair, adenine, guanine or genetic alphabet, checked by grep. (shorthand-evolution.html matches four letters and is about shorthand writing systems, which is a decoy and was read rather than counted.) The honest shape is small and good: a card, or a spread inside a piece about standards that turn out to be defaults. It is not a zine on its own, because the argument needs the biology to stay in the test tube and a zine would want it out.

Li, Q., Kim, H.-J., Liu, Y., Oh, J., Hou, P., Hoshika, S., Pintilie, G., Aiyer, S., Chong, J., Lyumkis, D., Benner, S. A. & Wang, D. “Structural basis of transcription of the hachimoji eight-letter alphabet by E. coli RNA polymerase.” Nature Communications 17, published online 2 September 2026. DOI 10.1038/s41467-026-76668-0 · PMID 42686745 · PMCID PMC13538358. UC San Diego, with the Foundation for Applied Molecular Evolution and Scripps. Read: the paper, in full, at PubMed Central, under CC BY 4.0 — including the methods, which is where the in vitro point above comes from. Not read: the UC San Diego release, or Hoshika et al. 2019. Wire: ScienceDaily, 4 September 2026.

Plausible Published online 2026-09-07 Journal BioScience Read the abstract, at Crossref

Lag 2 d

Most insects live in the tropics; most of what is known about insects does not

A global team assembled by Shawan Chowdhury at Monash names the gap plainly: insects are declining, they are largely absent from biodiversity conservation measures, and the gap is most severe in the tropics, which harbour over 80% of insect species. The abstract identifies four obstacles — limited global visibility, a data shortfall, a lack of infrastructure, and insufficient conservation action — and proposes a framework that integrates diverse data sources, strengthens regional expertise and links evidence to policy, worked through an application in Bangladesh. The point is not that nobody is looking at tropical insects. It is that when they do, the looking does not become visible knowledge: the release reports 111 relevant butterfly studies in Bangladesh with only 34 published in international journals, and Facebook holding records for 167 of 169 threatened Bangladeshi butterfly species.

The star stuff frame This is the How We Got Here motive without a chain attached: a body of knowledge that measures the surface, badly, and then acts as though the measurement were the world. A conservation map drawn from where the data happen to have been collected is a map of collectors, not of insects — and the missing eighty per cent are not missing because nobody knew them but because the people who knew them were not counted as knowing. Local expertise, museum drawers and a Facebook group are proposed here as data infrastructure, which is a quietly radical sentence about whose observation counts as observation.
A framework paper is a proposal, and the striking numbers are the release’s This is not a finding. It is an expert synthesis proposing a strategy, and every such paper is an argument about priorities as much as a description of a shortfall — the grade reflects that as much as the unread full text. Only two figures above are the paper’s own: the over 80% of insect species in the tropics, and the four named obstacles, both in the abstract we read. Everything else — 111 and 34 studies, 167 of 169 species, the 26% of GBIF insect records from the United Kingdom, the 14–30 million estimate of insect species, the 20% of diversity in temperate regions — comes from the Monash release and is marked as such rather than carried as the paper’s. Several of them appear to belong to a 2023 companion paper in One Earth that the release also cites, and we have opened neither. A digest that cannot say which paper a number is from has not traced that number.
Where it goes Epistemic injustice is well worked here already — grep finds the phrase on No. 39, No. 40, No. 60, No. 62, No. 70 and Six Principles and a Room. We have not re-read those six, so treat the overlap as unchecked rather than absent. What looks unspent is the specific move: not who gets believed but whose looking is infrastructure. That is a real distinction and it would need a better source than an unread framework paper before anyone builds on it.

Chowdhury, S., Jahan, S. F., Akite, P., Benedetti, Y., Bonebrake, T. C., Freitas, A. V. L., Hossain, M. M., Ladle, R. J., Morelli, F., Stork, N. E. & Kunte, K. “A framework for the strategic conservation of tropical insects.” BioScience, published online 7 September 2026. DOI 10.1093/biosci/biag136. CC BY 4.0 per the Crossref licence record. Read: the paper’s abstract, as deposited by the publisher at Crossref. Not read: the paper — academic.oup.com was refused at the network layer, and there is no PubMed Central copy — nor the Monash release beyond a summary of it, nor the 2023 One Earth companion (DOI 10.1016/j.oneear.2023.08.025). Not the paper’s own: every number in the finding above except the 80% and the four obstacles. Wire: ScienceDaily, 9 September 2026.

Verified Published online 2026-08-25 Journal PNAS Read the paper, in full

Lag 14 d

Nine species in one Jurassic forest, and no two of them sang at the same pitch

Insect wings fossilise, and so do the ridged files and scrapers they sing with. Working from 20 fossils of nine species — seven Prophalangopsidae and two Haglidae — all from the Jiulongshan Formation at Daohugou in Inner Mongolia, and all from one place and time (the paper gives ~165 million years, the assemblage 157–165), this group combined phylogenetics, laser Doppler vibrometry on living relatives, numerical simulation and a machine-learning model of syllable rate to reconstruct what the assemblage sounded like. Two results. The songs were pure tones, which the authors read as an adaptation to avoid being located by eavesdropping predators. And the assemblage showed, in the paper’s words, rich diversity in their calling song frequencies and repertoires, produced by differences in file morphology and wing size. One species called above 20 kHz — ultrasound, roughly 110 million years before bats appear in the Eocene, which the paper dates at about 55 Ma.

The star stuff frame A soundscape is not one sound. Nine species sharing a forest solved the same problem — be heard by the right listener, not by the wrong one — and arrived at nine answers, because a frequency that nobody else is using is worth more than a frequency that works well. The variation is not incidental to the system; it is the mechanism. Species recognition here is a difference-detection problem, and the diversity of the chorus is what makes any single voice legible. This site keeps arriving at that from the human end, and here it is in the Middle Jurassic, in a paper with no interest whatever in the argument.
Reconstruction, not recording — and they are not katydids Nothing here was heard. Every call is modelled, from wing geometry plus measurements on living relatives plus a phylogenetic prior, and the temporal pattern — the cadence — comes from a machine-learning model fitted on body size and Jurassic temperature estimates. The authors are explicit that the calls are inferred; a listener should hold the pitches more firmly than the rhythms. And the coverage’s word is wrong, gently. Science News calls these “katydid ancestors.” The paper says katydids (Tettigoniidae) are not known in the fossil record before the Cenozoic, and that molecular data put the family’s origin in the Late Cretaceous; the fossils here are Prophalangopsidae and Haglidae, whose nearest living relatives are the grigs. They are relatives of katydids’ ancestors, which is not the same claim and is the kind of compression that later gets quoted as a fact. We checked the ultrasound-versus-bats interval ourselves rather than repeating the coverage’s “about 100 million years”: 165 minus the paper’s own ~55 Ma for bats is about 110.
Where it goes Unspent, checked by grep: no page here contains stridulation, katydid or soundscape. This is the strongest More Than Human material to come through this scan in three editions, and the reason is that the umwelt argument is in the evidence rather than applied to it — the frequencies differ because the listeners differ. See the seed below.

Gu, J.-J., Montealegre-Z, F., Jonsson, T., Woodrow, C., Celiker, E., Islam, M. N., Linde, J. B., Sarria-S, F. A., Shi, F., Song, H., Robert, D. & Ren, D. “Reconstruction of an extinct soundscape reveals ultrasonic communication in the Jurassic.” PNAS 123(36), published online 25 August 2026, issue of 8 September 2026. DOI 10.1073/pnas.2615107123 · PMID 42640818 · PMCID PMC13552889. Read: the paper, in full, at PubMed Central, under CC BY 4.0. Not read: the supplementary appendix, or the audio reconstruction the paper supplies as Movie S4 and Audio S1 — so this entry has not heard the soundscape it is describing, which is worth saying out loud. PNAS’s own article page returned a 403. Wire: Science News, 8 September 2026; phys.org carried the same paper in late August, which we could not open.

Contested Published online 2026-06-11 Journal MNRAS Read the authors’ preprint, in full

Lag 88 d

Two papers, one journal volume, opposite conclusions about whether the universe is accelerating

The acceleration of cosmic expansion is measured from Type Ia supernovae, which are treated as standard candles. A known worry is that they are not quite standard: their brightness may drift with the age of the stellar population that produced them, and older populations sit at higher redshift. Sah, Rameez and Sarkar apply a redshift-dependent correction for progenitor age to the Pantheon+ catalogue — 1,701 supernovae, of which they cut 31 above redshift 0.8 — and report that this shifts the monopole component of the deceleration parameter q0 to positive values, meaning deceleration, while leaving a strong local dipole aligned with the bulk flow essentially unchanged. In the same volume of the same journal, Wiseman, Popovic, Sullivan, Riess, Scolnic, Chen, Davis and Galbany publish “Still accelerating: type Ia supernova cosmology is robust to host galaxy age evolution.” The two papers went online a day apart.

The star stuff frame Filed for the shape rather than the answer, and the shape is one this site trusts: a correction to the instrument proposed as an explanation for the result. Whether or not it survives, the argument is an argument about what a standard candle is standard against — and the honest position, which both sides hold, is that the measurement depends on a model of the thing measuring. That is not a weakness in cosmology; it is cosmology doing its job in public, with two papers in one volume and the disagreement printed rather than smoothed.
Why contested, and why the ScienceDaily headline is not the story “Dark energy debunked?” is a headline about one half of a standoff. This is a live, specialist disagreement between two teams — the second of which includes Adam Riess, who shared the Nobel Prize for the original acceleration measurement — published side by side, and a digest that reported only the first would be selecting for the surprising one. The grade is contested in the plainest sense the key allows. Two things we checked rather than assumed. The summary’s figure of “1,700+ supernovae” is not an inflation: the paper itself says 1,701 SNe Ia in Pantheon+, and we had drafted a correction claiming those were light curves before reading the paper and deleting it. And the tilted observer framing, which circulated with this story, does not appear in this paper; it belongs to the authors’ earlier work and to other coverage, and is not attributed to this result here. What we have not done is read the Wiseman paper at all, beyond its title and author list at Crossref — so this entry can report that a disagreement exists and cannot referee it, and does not try.
Where it goes Nowhere as a zine, and there is a specific reason to say so rather than leave it. Flying Free, the first Glimmers entry, is about the Roman Space Telescope launch, and one of the things Roman is built to measure is dark energy by Type Ia supernovae — that entry already carries an honest reading of the phrase, noting that NASA itself calls it the name astronomers gave to a mysterious something. That page is now the right place for a link, not a new piece, and the link should wait until the disagreement resolves or is shown not to.

Sah, A., Rameez, M. & Sarkar, S. “Pantheon+ supernovae corrected for progenitor age indicate the universe is decelerating.” Monthly Notices of the Royal Astronomical Society 549, 1–6, published online 11 June 2026. DOI 10.1093/mnras/stag844. TIFR Mumbai and Oxford. Accepted 30 April 2026; received in original form 16 March 2026. The other side: Wiseman, P., Popovic, B., Sullivan, M., Riess, A. G., Scolnic, D., Chen, R. C., Davis, T. M. & Galbany, L. “Still accelerating: type Ia supernova cosmology is robust to host galaxy age evolution.” MNRAS 549, published online 10 June 2026, DOI 10.1093/mnras/stag797 — not read. Read: the first paper in full as arXiv:2606.09650, posted 8 June 2026, the authors’ own text. Not read: the version of record — academic.oup.com refused — which is why this is not graded verified even though the text was read end to end. A Crossref oddity, left rather than tidied: this record gives a print date of 3 June 2026 and an online date of 11 June, in that order. We used the later, online date, which makes the lag a floor. Wire: ScienceDaily, 7 September 2026.

Verified Published online 2026-05-06 Journal Nature Read the paper, in full

Lag 123 d

A gut bacterium with one name turns out to be several populations living different lives

A species name in the gut microbiome behaves like an average nobody is. This group shows that genome-wide selective sweeps — one clone out-competing the rest in its niche, then rediversifying — are a pervasive mechanism sorting gut bacteria into distinct populations, in at least 66 taxa from 25 bacterial families. The resulting structure, in the paper’s phrase, is “akin to global epidemics across geographically and ethnically diverse human populations,” and the estimated divergence ages suggest sweep clusters can spread globally within decades. The clusters are ecologically real, not just phylogenetic bookkeeping: tested against five host phenotypes — colorectal cancer, Crohn’s disease, ulcerative colitis, type 2 diabetes and age over 65 — they associate differentially, some positively and some negatively. One Bacteroides uniformis cluster is shared between the Baka and Beti in Cameroon and the Matses in Peru, communities that have never been in contact; the paper reads this as a chain of transmission through less isolated populations, and offers two alternatives for why it is absent elsewhere.

The star stuff frame Two things, and the first is the one that carries. The taxonomic name was hiding the variation. What looked like one species doing one job is several populations that have each been selected into a different way of living — a spiky profile at the level of the genome, inside an organism we had been treating as a unit. Calling the whole thing by one name did not make it one thing; it made the differences invisible, which is the failure mode this site is organised around. And the second: a microbiome is partly received. Some of what is in you arrived from someone else, and the paper traces one lineage between two communities on different continents who never met. Interdependence at the resolution of a genome, and no version of it that is about individual virtue.
The word contagious is not in this paper The ScienceDaily headline is Your Gut Microbiome May Be More Contagious Than Scientists Thought. The word contagious does not occur once in the paper — we searched the full text; transmission occurs nine times, and the transmission claim is a single, carefully hedged inference about one B. uniformis cluster in three communities. The paper is about adaptation and population structure; the epidemic language is a simile for the shape of the phylogeny, not a claim about catching a microbiome from a housemate. A different 2026 paper does study strain sharing between cohabitants, and we did not open it. Two smaller things. The disease associations are associations, offered as evidence of ecological differentiation rather than of causation; the paper’s own suggestion is that sweep clusters may serve as markers, which is a diagnostic claim and not a therapeutic one. And the release says four health conditions where the paper tests five phenotypes — the abstract collapses Crohn’s and ulcerative colitis into inflammatory bowel diseases, and the analysis does not.
Where it goes Half spent, half not. You contain trillions of microbes and they are part of you is held by No. 7, which prints the 38 trillion to 30 trillion figure, and by the Symbioses entry The Collective You Carry. Unspent: that a bacterial species is not a unit either, and that the name is what made its populations look alike. That is a there is no standard member argument arriving one taxonomic level below where this site usually finds it, and it is the better half of the paper.

Yu, X. A., Strachan, C. R., Herbold, C. W., Lang, M., Gasche, C., Makristathis, A., Segata, N., Pollak, S., Tett, A. & Polz, M. F. “Genome-wide sweeps create ecological units in the human gut microbiome.” Nature 655(8121), 202–209, published online 6 May 2026, issue of 2 July 2026. DOI 10.1038/s41586-026-10476-w · PMID 42092154 · PMCID PMC13322978. University of Vienna, with the University of Veterinary Medicine Vienna and others. Read: the paper, in full, at PubMed Central, under CC BY 4.0. Not read: the Vienna release, or the cohabitation paper mentioned above. The lag axis undercounts this badly: the work was public as a bioRxiv preprint on 26 May 2024, DOI 10.1101/2024.05.25.595854, under the title “Genome-wide sweeps create fundamental ecological units in the human gut microbiome” — 834 days before the wire carried it here, against the 123 we print. That is the largest gap this page has recorded, ahead of the 753 days on last edition’s bats paper. Wire: ScienceDaily, 6 September 2026.

Plausible Published online 2026-03-06 Journal Astrophys. J. Lett. Read the authors’ preprint, in full

Lag 187 d

A comet from another system, carrying a chemistry ours mostly does not

3I/ATLAS is the third confirmed interstellar object to pass through the solar system. Using ALMA’s Atacama Compact Array on five dates in late 2025, this group detected methanol (CH3OH) and hydrogen cyanide (HCN) in its coma, across pre-perihelion distances of 2.6 to 1.7 astronomical units. The two molecules behaved differently: HCN production was depleted on the sunward side of the coma, while methanol was enhanced in that direction, and methanol production rose sharply from August to October with an uptick near the inner edge of the water sublimation zone. The headline number is a ratio, and it is a ratio: CH3OH/HCN of 124 (+30/−34) and 79 (+11/−14) on 12 and 15 September 2025 — among the most enriched values measured in any comet, and in the authors’ own words surpassed only by one comet of ours, the anomalous C/2016 R2 (Pan-STARRS).

The star stuff frame The atoms are the same atoms. What differs is the proportions, and the proportions are a record of the conditions the ice formed or was processed in — a different disc, a different star, a different temperature history. So an object from another system is not made of anything exotic; it is made of what we are made of, arranged by a different history. And it is not even off our chart: one comet born here sits further out on the same axis. The range is wider than the origin. That is a better sentence about difference than most things written about difference, and nobody wrote it about people.
A ratio is not an amount, and the headline is six months old “70 times more methanol than any comet from our Sun,” which circulated with this story, is wrong twice. The numbers are ratios of methanol to hydrogen cyanide, not multiples of any comet’s methanol; and the paper says explicitly that one solar-system comet exceeds them. The ScienceDaily framing — bursting with methanol, more than almost all known comets — is defensible; the multiplier version is not. The paper also declines a tidier claim than the coverage gives it. Statistical analysis put some methanol production in extended coma sources beyond 258 km at 99% confidence, but low signal-to-noise on the long baselines prevented definitively ruling out methanol as a pure parent species — so where the methanol comes from is open, and the paper says so. And this is not this week’s news. It was published on 6 March 2026 and covered then; a feed carried it to us on 9 September as though it were new. The lag bar at the bottom of this list is that, and it is the longest the axis has measured.
Where it goes Unspent as a subject, checked by grep: no page here says methanol, and none says ‘Oumuamua, Borisov or 3I/ATLAS. (Interstellar appears on eight pages, The Cloud Phase and Bodies of Water among them; we have not re-read them, so the overlap is unchecked rather than absent.) It should not be built on this entry. A piece about an interstellar visitor’s chemistry needs the version of record and needs the comparison comet handled properly, and the frame above is doing more work than a preprint abstract can carry alone.

Roth, N. X., Cordiner, M. A., Bockelée-Morvan, D., Biver, N., Crovisier, J., Milam, S. N., Lellouch, E., Santos-Sanz, P., Lis, D. C., Qi, C., Foster, K. D., Boissier, J., Furuya, K., Moreno, R., Charnley, S. B., Remijan, A. J., Kuan, Y.-J. & Hart, L. X. “CH3OH and HCN in Interstellar Comet 3I/ATLAS Mapped with the ALMA Atacama Compact Array: Distinct Outgassing Behaviors and a Remarkably High CH3OH/HCN Production Rate Ratio.” The Astrophysical Journal Letters 999(2), L32, published online 6 March 2026. DOI 10.3847/2041-8213/ae433b. Read: the published abstract, as deposited by the publisher at Crossref; and the authors’ preprint in fullarXiv:2511.20845, v1 posted 25 November 2025, v2 revised 2 March 2026, four days before journal publication. Not read: the version of record — iopscience.iop.org was refused at the network layer and doi.org returned a 403 — nor the NRAO or ALMA releases. Not graded verified for that reason. And the lag axis undercounts here too: public as a preprint 288 days before the wire, against the 187 printed. Wire: ScienceDaily, 9 September 2026.

Held

Five this week. The first is the one that earns the section outright, because it is not a story we declined for being thin — it is an error this page corrected last week, still circulating on the wire we scan. The second is a four-year-old paper arriving as a discovery. Nothing here was invented to fill the section, and nothing that was looked at and declined has been quietly omitted.

Held — already filed, and the coverage still carries the figure we corrected

More than 500 vesper bat species have two antibody gene sets

We filed the immunoglobulin duplication paper on 4 September, read it in full, graded it verified, and corrected exactly one figure: the widely repeated more than 500 vesper bat species. The paper measures 26 species with duplicated heavy-chain loci and confirms both loci functional in one, the big brown bat; it calls Vespertilionidae the largest bat family and gives that family no number at all, printing only “over 1500 known species of bats” for the whole order. The release that surfaced on this week’s scan still says more than 500 species of vesper bats. Held, obviously, because it is already filed — but recorded here because a correction that lives only on our page is a correction that has not happened, and because this is what the gap between a paper and its coverage looks like when you watch the same item twice. Where the release is right: its per-locus gene counts check out — we went back to the paper, and A-IGH does encode 33 IGHV genes against B-IGH’s 99. The error is the species number, not the reporting in general, and saying so is part of naming it fairly.

Held — a 2022 paper presented as this week’s discovery, plus a supplement pitch and mice

Stanford scientists discover a seafood that can reverse signs of aging

Sea squirts are rich in plasmalogens, a class of membrane lipid that declines with age; supplemented aged mice did better on learning and memory tests, with reduced neuroinflammation. Declined three times over. The paper is from 2022. Gu et al., Frontiers in Molecular Biosciences 9, published online 23 February 2022, DOI 10.3389/fmolb.2022.815320 — confirmed at Crossref, which is the only part of this item we traced. The wire carried it on 2 September 2026, 1,652 days later, with a headline in the present perfect. The attribution is also off: the author list is led from Shanghai Jiao Tong University, with one Stanford co-author, so Stanford scientists discover misplaces both the verb and the institution. And the framing is the trap this collection names — a mouse supplement result, written as a seafood you could go and buy. A finding does not become false by being four years old. It becomes a different kind of claim when it is presented as news, and we are not passing that on.

Held — a profile of a research programme, not a finding of the week

Winter work shows it really pays to be a smart mountain chickadee

Mountain chickadees cache tens of thousands of seeds and recover them by spatial memory through a Sierra winter; a long-running programme at Nevada, Reno has tracked how that ability varies and what it costs and buys. The Science News piece of 9 September is a feature about that programme, and the headline result it rests on — birds with better spatial cognition living longer — is a Science paper from 2024, which we did not open. So there is no paper of the week here to file. It is held on a second ground too, and that one is editorial. The available framing is smarter birds do better, which is a fine description of one selection pressure in one environment and reads, transplanted, as exactly the merit argument this collection refuses. The interesting version is the opposite: variation in spatial cognition persists in these populations despite that pressure, which is a question about what else the variation is buying. That would need the primary literature and a lot more care, and it is a better piece than the one on offer.

Held — two papers under one headline, neither openable

The hidden instructions that build the human brain

Radial glia are the stem cells that build the human cerebral cortex, and two UCLA-led papers reported this week describe what steers them: a metabolic route through the pentose phosphate pathway, where restricting glucose shifts which cell types get made, and physical contact with incoming thalamic signals, which alters the production of upper-layer neurons. This is squarely on-frame — development and timing, cortical cell-type composition set by conditions rather than by a plan. It is held because the coverage folds two separate papers, in Cell and in Science, into a single headline, and a digest that cannot tell you which finding belongs to which paper has not traced either. Both publishers refused, and the UCLA newsroom was refused at the network layer as well, so we could not separate them. Carried here as the most promising thing we could not open, and a first candidate for next week.

Held — observational associations converted into personal advice

Semaglutide and psychiatric outcomes, early statins and dementia, blood tyrosine and male lifespan

Three items from the same week’s mind-and-brain feed, held together because they share one failure and it is not a failure of the underlying work. Each is a large observational study — a Swedish cohort of about 15,000 people with bipolar disorder, a Danish cohort of more than 132,000 with type 2 diabetes, a genetic-and-observational analysis of more than 270,000 — and each reaches the reader as a percentage attached to a decision they might make. None of the three named its journal in what we could reach, so the sourcing floor is a summary, and that alone holds them. But the editorial reason is the one worth writing down: a hazard ratio drawn from who happened to be prescribed what is a fact about a health system, and printing it as a number a person can act on moves a question about access and prescribing into an individual’s morning routine. That is the same conversion this collection declined on the sitting-and-cancer item two editions ago, and declining it twice is the point of keeping the list.

Seedbed

Two seeds, both from items read in full, which is the first time this has been true of every seed on the page. Plus the leads nobody opened.

Zine · More Than Human

Nine ways to be heard in one forest

The strongest thing to come through this scan in three editions, and it can be built from a single paper that has been read end to end. Nine species in one Middle Jurassic forest, each singing at its own frequency, the pitches set by file morphology and wing size, the tones kept pure so that predators could not locate them — and one of the nine already up in the ultrasound, a hundred and ten million years before the bats that would eventually listen there. The argument is that the diversity is the mechanism, not a by-product of it: a call only identifies a species if the other calls are different, so difference is what makes any individual voice legible. The reading a zine would have to earn, and could: the same is true of us, and it is the reason a norm is a worse tool than a range. What is missing before this is a piece: the supplementary material and the authors’ own audio reconstruction, neither of which we have opened — and a piece about a soundscape that has not heard the soundscape would be writing about a diagram.

From the PNAS paper filed above, read in full at PubMed Central. Not obtained: SI Appendix, Movie S4, Audio S1.

Field guide entry · or a spread

There is no standard member of a species

This site’s field guides keep arriving at there is no standard star, nest, egg, shark, rainbow, bird, way to rest. The gut-sweeps paper puts the same claim one level lower, where it is stranger: a bacterial species name covers several populations that have been selected into different ways of living, in at least 66 taxa across 25 families — and the name is what made them look alike. The transferable point is about naming rather than about bacteria. A taxon is a claim that a set of things are the same kind of thing, and the finding is that the claim was hiding the structure it was supposed to describe. What holds this at seed: one paper, however well read, is one paper, and the argument needs a second case from a different kingdom before it is a pattern rather than a good line. It also needs the existing count re-derived rather than quoted — last edition put the guides arriving at that formula at fifteen, and the standing lesson in our own working notes is that every such figure is a lead and not a fact.

From the Nature paper filed above, read in full at PubMed Central, plus you-were-never-one-thing-zine.html and symbioses-field-guide.html, read in the repository. Not obtained: a second case.

Leads · unchecked

Four off this week’s wires

Newly reported Denisovan fossils from a Chinese cave site. Morning glories reported to have lost adaptive capacity over nine years under climate change and pollinator scarcity, which is the inverse of the usual evolution-is-fast story and would need handling with care. Engineered bacterial circuits at MIT, described as living transistors. And a survey of several hundred astrobiologists in which only 6.6% found the K2-18b biosignature claim convincing — which is an epistemics item rather than a discovery, and possibly the most useful of the four: a field publishing its own distribution of belief instead of its loudest member.

All four off ScienceDaily and Science News listings, 4–9 September 2026, and none of them opened — no journal, authors, date or DOI obtained for any. The 6.6% is a summary’s figure and is not confirmed against the survey. Listed so that next week starts from somewhere. Last edition’s three leads were not chased either, and are not carried forward here; they are still in that edition.

Not

Not a citation. Three papers here were read in full and a fourth as its preprint, which is the best this page has managed — and it changes nothing about the standing rule. Nothing filed here may be promoted to a zine, a Glimmers entry or a card without the primary source in hand, a changelog entry and its row in the fact-check ledger. For the three items resting on abstracts, lifting a figure out and printing it elsewhere would be laundering a deposit into a fact.

Not a superpower. Twenty-six bat species, nine Jurassic singers, several populations inside one bacterial name: these are facts about architecture and range, never about merit. The chickadee item is held partly because its available framing runs the other way, and better spatial memory buying a longer life in one Sierra winter is a description of a pressure, not a ranking of birds.

Not a cure list. The sea-squirt item is held for going that way, and the three observational items with it. A mouse, a supplement and a percentage are three different things, and a headline that joins them is selling something.

Not anti-science. The cosmology item files a live disagreement between two teams and does not referee it; the complaint is about a headline that reported half a standoff. If a draft off this digest starts reading as a case against evidence, or against the people making measurements this careful, it has gone wrong.

Not “science says you are valid.” A finding that makes disbelieved people credible also makes them revocable. The belonging claim was never waiting on a result, and it does not move when one fails to replicate — or when two papers in one journal volume disagree about the universe.

Not because of. A comet from another star carrying the same atoms in different proportions is a rhyme, not a proof. The universe declining to pathologize its own variation is why the phrase carries — down to your star stuff measures how far the loving reaches, never why it is owed.