---
title: "Glimmer Wire"
url: "https://starstuff.earth/glimmer-wire-2026-10-02.html"
updated: "2026-10-02"
description: "Edition six of Glimmer Wire, 2 October 2026 — the weekly graded scan behind Glimmers. Eight items filed, six held, six papers read end to end, median lag 20 days. Sharks that hear far past the limit set for them by an organ they do not have, eleventh-century children buried with adults who were almost never their kin, a rare syndrome whose old prevalence figure was a measurement of who got tested, and a scroll built and burned on purpose so an algorithm could be marked against a known answer."
collection: "Glimmer Wire"
licence: "CC-BY-SA-4.0"
licence_url: "https://creativecommons.org/licenses/by-sa/4.0/"
fact_check: "https://github.com/Stimpunks/Star-Stuff/blob/main/FACTCHECK.md"
generated_by: "tools/build-markdown.mjs, from the page's own <main> landmark"
---

[Stimpunks](https://stimpunks.org/) × [More Realms](https://morerealms.com/) · Glimmer ★ Wire · Edition Six

[Stimpunks Foundation](https://stimpunks.org/) × [More Realms](https://morerealms.com/) · Working Paper

# Edition of 2 October 2026

A weekly read of what is going well in science and medicine, graded before it is enjoyed.

The sixth edition of [Glimmer Wire](https://starstuff.earth/collection-glimmer-wire.html), the scan that sits behind [Glimmers](https://starstuff.earth/collection-glimmers.html). **Six papers were read end to end**, all six of them versions of record — the best week this page has had, and the reason is one open-access repository answering on the second attempt rather than any new skill of ours. Items are sorted by *lag*, which runs from **4 days to 91**, median **20**: less than a quarter of last week’s, because four of the eight reached us inside ten days. **Five are graded verified, one contested, two plausible.** The section that matters most is [Held](#held-items-1002), which lists what we declined and why, and this week it contains a kind of refusal this page has not had to write before: **an index offered us a summary written by a language model, in the slot where an abstract goes, correctly labelled and entirely convincing.**

---

*Edition* **Six** *Published* **2 October 2026** *Amended* **—** *Verified* **5 of 8**

8 filed · 6 held · 4 seeds · 5 verified · median lag 20 d

## What was actually read: six papers of record in full, four publisher deposits, and one summary written by a machine

**Answered when tried, by name:** api.crossref.org, api.semanticscholar.org, pmc.ncbi.nlm.nih.gov (on a retry, four times out of four), journals.plos.org, arxiv.org and sciencedaily.com. **Refused at the network layer — the connection never completed, which is not a paywall and is recorded as what it was:** www.nature.com, www.frontiersin.org, academic.oup.com, onlinelibrary.wiley.com, doi.org, export.arxiv.org, www.ncbi.nlm.nih.gov, europepmc.org, www.ebi.ac.uk, api.openalex.org and api.unpaywall.org. **Refused by the publisher, with a real answer:** www.science.org (HTTP 403). **Rate-limited:** www.biorxiv.org (HTTP 429, a 17-byte body). *Three of those refusals sit in front of papers that are openly licensed* — the stick-insect revision is CC BY at Wiley, the turtle tracking is CC BY at Frontiers, and the bat phylogeny is CC BY at Nature. An open licence is a permission, not a road.

**Two host behaviours from last week repeated exactly, and one is worth the byte count again.** A wrong search path at PubMed Central still returns the PMC home page at HTTP 200 — **67,236 bytes, the identical figure recorded on 25 September**, delivered four times over for four different queries. A week apart, two different sets of search terms, the same byte count to the digit: that is the tell, and the status code is not. And PMC again served Google’s reCAPTCHA challenge at 200 — a 20,420-byte document whose first line is a \<base\> pointing at google.com/recaptcha — for every article requested on the first attempt. Retrying with a different user agent returned all four papers in full. The right route was open the whole time; the first knock was answered by a doorman.

**The new one is worse than either, and it is the reason this edition holds a story it wanted.** Where a publisher has elided an abstract, Semantic Scholar’s API returns "abstract": null and, beside it, a tldr field containing fluent, specific, paragraph-shaped prose about the paper. It is generated by a language model; the API says so, naming the model in the same object. For the *Nature* paper on zhelestid mammals that text was the only summary available anywhere we could reach, and it reads exactly like an abstract. **It is not a source, and nothing about the response would stop a tool — or a person skimming — from treating it as one.** The field is honestly labelled, which is to the archive’s credit and is the entire defence; the laundering here would be structural rather than anybody’s fault. *A status code is not a source, a byte count is not a page, and now: a well-formed field in the right place is not an abstract.* That paper is [held](#held-zhelestid-1002).

**What was bought is six papers read end to end, all six versions of record.** Four at PubMed Central — the *Science Advances* paper on Enceladus’ ice grains, the *Science Advances* paper on children in early Christian burials, the *Autism Research* paper on Phelan–McDermid prevalence, and the *Integrative Organismal Biology* paper on blacktip sharks. Two at the publisher, openly licensed and open in fact — the *PLOS One* model-scroll paper and the *PLOS Biology* lactoferrin paper, the second of which is in the [Seedbed](#seedbed-1002) rather than filed. **The remaining four filed and five held items rest on publisher-deposited abstracts at Crossref**, which in every case here is the complete structured abstract and is still a deposit the publisher wrote, one link out from the paper.

**The median lag fell hard — 20 days against 85 last week — and the distribution is the same two populations as ever.** Four items arrived inside ten days (4, 6, 9, 9) and four had been sitting for a month to three (31, 77, 88, 91). Nothing lands in between, which is the third consecutive edition with that shape. **Two lags are measured from an issue date rather than an online-first date and are marked where they appear** — Crossref deposits no published-online value for either *Science Advances* paper — which makes those two figures floors rather than measurements. *No preprint was involved in any filed item this week*, so the axis’s standing blind spot had nothing to hide; that is an absence worth recording rather than a clean bill.

**A note on the wire dates.** ScienceDaily’s indexes again list several of these under a date one day later than the timestamp inside the release’s own URL. Every lag below is computed against the *URL timestamp*, and every entry prints the wire date it used.

## Filed

Sorted by lag, shortest first. Bars are scaled to the longest item, 91 days. **Four of the eight reached us within ten days of publication** — lags of 4, 6, 9 and 9 — which is the fastest half this page has filed. The other four had been in their journals between **31 and 91 days** before a feed handed them over this week as news.

Verified Primary source read and confirmed.

Plausible Finding sound and metadata confirmed; version of record not read.

Contested Specialists disagree, or the claim outruns the method.

Held Could not be traced to a source that supports it.

Verified *Published* 2026-09-25 (issue) *Journal* Science Advances *Read* the paper, in full

Lag 4 d

### Every grain of Enceladus is a different ocean, and the difference is made on the way out

Enceladus sprays its subsurface ocean into space, and Cassini’s Cosmic Dust Analyzer spent thirteen years tasting the spray. The salt-rich grains — Type 3 — have always been read as frozen droplets of ocean water, so the expectation was that they would all taste much the same. Across **961 individual mass spectra** they do not: the authors find **at least five compositional subtypes**, dominated variously by sodium chloride, sodium carbonate or bicarbonate, sodium phosphate, sodium hydroxide, potassium chloride or potassium hydroxide — and in sodium-rich grains, chloride features and carbonate features are *almost never present together*. Freezing experiments and thermodynamic modelling supply the mechanism, and it is entirely about how the sample was taken. Different salts precipitate at different temperatures: phosphates first at around −1 °C, carbonates from about −4 °C, chlorides and hydroxides last, near −23 °C. In a droplet that freezes *slowly* — the paper’s condition is below 20 kelvins a minute, in droplets of tens to hundreds of micrometres — those events are separated in time, so the salts end up separated in space inside one grain. The droplets are then accelerated past 100 metres a second up narrow ice vents, collide repeatedly with the walls, and shatter. Each fragment that reaches the instrument carries mostly whichever salt happened to occupy its corner of the parent drop.

**The star stuff frame** **Not one of those grains is a sample of the ocean, and the variety between them is not noise around the ocean’s composition — it is manufactured by the act of sampling.** The paper says it plainly: the composition of each subtype “is inherited from the composition of regions within the parent grains.” Read a single grain and you learn about a corner of one droplet and the rate at which it froze. What makes this worth a page here is not the moon; it is that the same sentence is true of the [prevalence figure](#prevalence-1002) four items down, where a number everybody quoted about a syndrome turned out to describe the route to a genetic test. *Before you can read a measurement as a fact about its subject, you have to model the thing that selected it.* One of those two cases is ice and the other is insurance criteria, and the error is identical.

**What is careful about it, and the two numbers** **The abstract makes no habitability claim at all**, which on an Enceladus paper in 2026 is a decision somebody took, and it deserves naming where this page usually names the opposite. The phosphates are there, the pH modelling is there, the words about life are not. Two numbers to keep straight, both the paper’s own: the abstract says **“≈1000”** spectra and the results work with **961**; and the abstract’s freezing condition, “below 20 kelvins per minute,” is the outer bound, while the discussion says only rates “in the order of 10 K/min” produced the segregation actually observed. The authors are also explicit that some of the data is not yet reproduced by their experiments — the degree of sodium segregation, and the high potassium in the potassium-rich group. **The lag is a floor, not a measurement:** Crossref deposits no online-first date for this paper, only the 25 September issue, so 4 days is the shortest the true interval can be.

**Where it goes** Checked against the site as it stood before this page existed. Enceladus already appears on [*A Field Guide to Worlds That Work by Other Rules*](https://starstuff.earth/worlds-field-guide.html), so the moon is spent. The *mechanism* is not: bubble bursting and ascertainment return nothing here, and ascertain appears once, in a sourcing note on [No. 1](https://starstuff.earth/bone-song-zine.html). The piece this wants to be is a [Star Stuff](https://starstuff.earth/collection-star-stuff.html) zine about the sampling instrument rather than the moon, with the prevalence item as its second panel — one settled fact, followed until the belonging claim is already inside it.

**Postberg, F., Zou, Z., Sekine, Y., Koga, M., Schmidt, J., Fox-Powell, M., Klenner, F., Hillier, J. K., Khawaja, N., Kadono, T., &Ccedil;akar, M., Kempf, S. & Srama, R.** “Cassini CDA observes compositional segregation of Enceladus’ ice grains from slow freezing and fragmentation of oceanic spray.” *Science Advances* **12**(39), issue of 25 September 2026. **DOI** 10.1126/sciadv.aee7256. **Read:** the published article in full at PubMed Central (PMC13614379), including the discussion and the experimental-limitations section; every quotation and figure above is from it. **Not read:** the Supplementary Materials (Tables S1–S7, Figs. S1–S13), which carry the statistical significance of the salt separation and the vent-condition modelling. www.science.org returns 403. **Date note:** issue date only; the 4-day lag is a floor. **Wire:** ScienceDaily, 29 September 2026.

Contested *Published online* 2026-09-23 *Journal* Nature *Read* the abstract, deposited at Crossref

Lag 6 d

### A new bat family tree, and a paper that explains why the old answers were wrong rather than why the old people were

The Bat1K consortium assembled **chromosome-level genomes for 103 bat species — 42 of them new — covering all 21 bat families**, and combined them with a morphological dataset of **699 characters for 65 species, including 44 pre-Quaternary fossils**. The resulting phylogeny moves Myzopodidae to the earliest branch within Vespertilionoidea and makes Emballonuroidea and Vespertilionoidea sister groups. Two results travel beyond systematics. First, the biogeography: fossilised birth–death and dispersal–extinction–cladogenesis analyses place the origin of bats, “and thus powered flight,” in **Europe in the late Palaeocene**, which the abstract says is “refuting African and North American origins.” Second, the placement of the fossil *†Vielasia* in the oldest ‘Eochiroptera’ clade “indicates that laryngeal echolocation predates crown-bat diversification” — echolocation before the living families split, rather than after. The sentence this page keeps is a smaller one: the analysis “revealed a mosaic evolutionary history across bats and explained why previous phylogenetic studies were misled.”

**The star stuff frame** *Explained why previous studies were misled.* Not corrected, not refuted — **diagnosed**. A mosaic history, where different parts of the genome carry different trees, will hand a consistent and consistently wrong answer to anyone sampling too little of it, and every earlier team was doing the honest thing with the data that existed. That is the [How We Got Here](https://starstuff.earth/collection-how-we-got-here.html) move: the chain does not end at *they were wrong*, it ends at *here is the property of the world that made the wrong answer look right*. The site already owns the general case — [*Five Sigma*](https://starstuff.earth/five-sigma-zine.html), No. 40, on a threshold nobody re-examined — and this is the version where the shortcut was not a convenience but a sampling limit, which is the more common and less blameable kind.

**Why contested, and the number the headline invented** **A continent of origin inferred from fossil-calibrated models is partly a statement about where fossils have been collected and prepared.** The Palaeogene bat record of Europe is the best-sampled in the world, and the paper’s own pivotal fossil, *†Vielasia*, is European. The authors hedge accordingly — bats “*probably* originated in Europe” — and biogeographic reconstruction from an uneven fossil record is exactly the kind of result specialists argue about for a decade. We have not read the paper and cannot say how its sampling bias is handled; the abstract does not address it. **And one figure in the coverage belongs to nobody but the headline writer.** “Bats may have first evolved in Europe *65 million years ago*” — **65 Ma appears neither in the abstract nor in the university’s own release**, both of which say *late Palaeocene*. The late Palaeocene is conventionally about 59.2 to 56.0 million years ago. The headline is therefore roughly six million years early and puts the origin of bats at the Cretaceous–Palaeogene boundary rather than several million years after it, which is a different claim about a different world. Separately: the release says “137 researchers from 64 countries” where Crossref’s deposit lists **136 authors**. Those need not conflict — a consortium has contributors who are not authors — but only one of the two numbers is checkable, and it is the smaller one.

**Where it goes** Bats are on this site already and in one register only: [*A Field Guide to Resting*](https://starstuff.earth/rest-field-guide.html) uses “the famous twenty-hour bat” as a *sourcing* caution — one 1960s measurement at one ambient temperature, which halves in a cooler room. Nothing here touches bat phylogeny, biogeography or the origin of flight; echolocation appears on three pages, all of them about what an animal does rather than when the ability arose. The unspent argument is the diagnosis: *why a method produced a stable wrong answer*. Hold it until the paper can be read — a chain built on an abstract would be a chain whose first joint is contested and whose evidence is a deposit.

**Morales, A. E., Liang, Y., Thomas, W. R., Leushkin, E. V., Castellanos, F. X. *et al.*** (136 authors; Teeling, E. C., last). “Reference genomes and fossils revise bat family phylogeny and biogeography.” *Nature* **658**(8134), 141–152, online 23 September 2026. **DOI** 10.1038/s41586-026-11007-3. CC BY 4.0. **Read:** the publisher-deposited abstract at Crossref, in full; every quotation above is verbatim from it. The “137 researchers from 64 countries” figure and the dispersal narrative are from University College Dublin’s release and are marked as the release’s. **Not read:** the paper, which is CC BY and which we could not reach — www.nature.com is refused at the network layer and no PubMed Central record exists. **Wire:** ScienceDaily, 29 September 2026.

Plausible *Published online* 2026-09-21 *Journal* Austral Entomology *Read* the abstract, deposited at Crossref

Lag 9 d

### One paper makes two species, raises one from the dead, buries a name, and nails three more to particular insects in drawers

A revision of the Australian members of the stick-insect genus *Anchiale* takes the group from two recognised species to **five**, and the five directions the names move in are the interesting part. *A. mabiensis* and *A. robusta* are described as new. *A. albopunctata* (Kirby, 1896) is **removed from synonymy** with *A. briareus* and restored as a species in its own right — stat. rev., a name published in 1896 and brought back out of another one, after a spell as a synonym whose start date the abstract does not give. Going the other way, *Ctenomorpha tessulata* Gray, 1835 is sunk as a **new synonym** of *A. briareus*, taking with it *A. austrotessulata*, the replacement name coined for it in 2007. And **lectotypes are designated for three names** — which is the quietest act in the paper and the most load-bearing: it fixes each name permanently to one chosen specimen, so that the next revision argues about animals rather than about what somebody meant. The decisions are supported by Bayesian and maximum-likelihood phylogenies on a mitochondrial COXI fragment, for every species except *A. spinicollis*. The paper closes with a dichotomous key to all five species *and their eggs*.

**The star stuff frame** Last week this scan filed a penguin described in 2026 from a skin collected on **31 December 1897**, in a paper that simultaneously demoted a species another team had named in 2020, and the [seed it produced](https://starstuff.earth/glimmer-wire-2026-09-25.html#seed-drawer-0925) was that *a name is a decision somebody makes about where a boundary goes, and it moves in both directions, and the animal is unaffected either way.* Seven days later, a second paper in a different phylum does all four motions at once. **Two cases is the difference between an observation and a pattern**, and this is the seed earning its second leg. For a reader who has been handed a diagnosis, a label or a category as though it were a discovery about them, the transferable fact is not that categories are arbitrary — these are evidence-led, careful, and better than what they replace — but that **they have dates, authors, and a procedure for changing**, and the organism goes on exactly as it was through every revision.

**What the evidence is, and what the coverage added** **The molecular support is one mitochondrial gene fragment, and one of the five species was not sequenced at all.** Species delimitation on COXI alone is the most common and the weakest widely used form of this evidence — mitochondria are inherited as a single unit and can tell a different story from the rest of the genome — and the paper is open about which taxon is missing. That is why this is graded plausible on its own terms as well as on ours; the abstract also makes clear the decisions rest on morphology and type material as much as on the tree. Separately, most of what the coverage says is not from this paper: **the 62 cm Australian record, the 3,609 species worldwide, the ant-mimicking nymphs of *Extatosoma* and the leg-shedding are a fun-facts block in the release**, not findings here, and none of them is graded on this page. The 1,000-plus iNaturalist and Atlas of Living Australia observations *are* part of the story — the release’s point is that a heavily photographed insect went unrecognised anyway — but we have that from the release, not the abstract.

**Where it goes** lectotype, stick insect and Phasmid each return zero across this site; synonym hits six pages, all of them using the word about words rather than about taxonomy. So the nomenclatural machinery is entirely unspent, and it is the half with the argument in it. This belongs with the penguin as *one* piece and not two — a [How We Got Here](https://starstuff.earth/collection-how-we-got-here.html) chain through how a name is made, unmade and pinned, with the first joint marked documented because both papers document their own acts in the formal vocabulary of the code.

**Jones, B. R., Brock, P. D., Coupland, R. M., Emmott, A. & Lo, N.** “A Revision of Australian *Anchiale* Stål, 1875 (Phasmida: Phasmatidae), with the Description of Two New Species.” *Austral Entomology* **65**(4), online 21 September 2026. **DOI** 10.1111/aen.70096. CC BY 4.0. **Read:** the publisher-deposited abstract at Crossref, in full — a complete taxonomic abstract naming every act; every nomenclatural statement above is from it. **Not read:** the paper, the key, the egg descriptions and the phylogenetic figures. The article is CC BY and onlinelibrary.wiley.com is refused at the network layer, which is a refusal of the road rather than of the licence. **Wire:** ScienceDaily, 30 September 2026.

Verified *Published online* 2026-09-16 *Journal* PLOS One *Read* the paper, in full

Lag 9 d

### They wrote a scroll, carbonised it, and then read it back — because the only way to grade an algorithm is to know the answer first

The Herculaneum papyri were turned to charcoal by Vesuvius in 79 CE and are being read, slowly, by X-ray tomography and machine learning. The obstacle is that carbon ink on carbonised papyrus is nearly invisible to X-rays: both are carbon, so there is almost nothing to see. Lead in the ink would change that — lead has been found in two Herculaneum fragments — but you cannot test the idea on the scrolls, because they are irreplaceable and synchrotron time is scarce. So this team **built one**. Modern Egyptian papyrus; carbon ink mixed with a soluble lead salt at concentrations varied line by line; text written by hand with a stylus; rolled into a cylinder 40 mm across and 110 mm long; then **carbonised in a semi-sealed steel container in a furnace with little or no oxygen**, which produced an object the authors say closely resembles a real Herculaneum scroll, “with tight packing of layers and waviness that will challenge unrolling algorithms.” Imaged on a laboratory X-ray tomography system at NIST, virtually unrolled with their own software, and read. The leaded text came up. The two lines written *without* lead were, as predicted, invisible — the control doing its job. And a handheld XRF device detected the lead straight through the carbonisation, which suggests a cheap way to screen which real scrolls are worth the beam time.

**The star stuff frame** **This site has, from its first week, named the discipline that gets dropped — and has never once been able to point at it being kept.** [No. 40](https://starstuff.earth/five-sigma-zine.html) is about a threshold chosen for convenience; [*Too Good to Check*](https://starstuff.earth/too-good-to-check.html) is a catalogue of how a fact goes wrong here, our own facts included. All corrections, no picture of the alternative. Here is one, and the whole of it is in a single clause of the discussion: *“Since that text was initially written on the scroll before carbonization…”* — they know the answer. The object exists so that an algorithm’s output can be **marked** rather than admired, and so that failure is legible. That is the same move as last week’s gravity key in a sealed envelope and the fortnight-old survey that published the spread of physicists’ beliefs instead of a spokesperson’s: in all three, **the people producing the result arranged in advance not to be able to steer it.** This is the third case and the first outside physics, which is what the [Notes-page seed](#seed-groundtruth-1002) was waiting for.

**What it is not, and what the headline got backwards** **It is a proof of concept on one model scroll with a modern, soluble lead salt.** How lead got into ancient ink is, in the paper’s own words, still “under investigation,” with the ink itself and drying additives both candidates — so a model made by dissolving lead nitrate into carbon ink reproduces the *X-ray consequence* of ancient leaded ink without claiming to reproduce its chemistry. Nothing here reads a real scroll, and the imaging was a laboratory system, not a synchrotron. The authors also apply no text-recognition algorithm at all: the fragment images were flat-field corrected and contrast-adjusted, nothing more, which is honest and also means the paper demonstrates contrast rather than recovery. **And the coverage has the physics backwards in its verb.** “Scientists burned their own scrolls” — they *carbonised* them, in a sealed container with the oxygen excluded, which is close to the opposite of burning and is precisely the reason anything survives at Herculaneum to argue about.

**Where it goes** ground truth appears exactly once on this site, on [*The Difference-First Frame*](https://starstuff.earth/difference-first-frame.html), and in the opposite sense: a diagnostic model measures us against a typical profile “that was never the ground truth to begin with.” That is a *collision* rather than a duplication, and it is the better pairing — one page about a ground truth that does not exist and never did, one case of somebody building one on purpose because the work was impossible without it. Herculaneum, papyrus, blinding and pre-registration all return zero here. The piece is a [Notes & Rationale](https://starstuff.earth/collection-notes.html) page, not a zine; see the seed below for why it still does not have a title.

**Seiler, D., LaManna, J. M., McOsker, M., Kreimer, D., Daugherty, M. C. & Dopke, J.** “A model carbonized papyrus scroll opens a novel path to identifying readable scrolls of the Herculaneum Library.” *PLOS One* **21**(9), e0353485, online 16 September 2026. **DOI** 10.1371/journal.pone.0353485. **CC0** public domain dedication. **Read:** the paper in full at journals.plos.org, including methods, figure captions and discussion; every quotation above is from it. Received 14 March 2026, accepted 24 June, published 16 September — the authors’ own dates, printed because this page is about intervals. **Not read:** the deposited dataset at doi.org/10.18434/mds2-4227. **Wire:** ScienceDaily, 25 September 2026.

Plausible *Published online* 2026-08-27 *Journal* Frontiers in Amphibian and Reptile Science *Read* the abstract, deposited at Crossref

Lag 31 d

### Nine turtles, and a home range whose spread is twice its own average

Two genetically distinct forms of green turtle share the water in the Galápagos and nowhere else: the Central Eastern Pacific “black morphotype” and the putative Western Indo-Pacific “yellow morphotype.” What the yellow-morph animals do there had never been tracked. Nine adults — **two females and seven males** — were fitted with Argos satellite tags across four islands and followed for **between 45 and 418 days**. They stayed. Core areas sat in shallow nearshore habitat, most individuals kept to a single island, several crossed between islands over tens of kilometres, and **not one left the archipelago** for the length of its tag. The sentence this page keeps is the one about spread: home ranges were “highly variable among individuals (mean = 1,497.9 ± 2,913.9 km² SE).”

**The star stuff frame** **The dispersion is roughly twice the mean.** Whatever that figure is — and see below, because its label is ambiguous — a single average home range for this group describes none of the nine animals in it; it is a number produced by nine quite different lives and resembling no one of them. [*A Field Guide to Being a Turtle*](https://starstuff.earth/turtle-field-guide.html) already says, of the order as a whole, that something this varied *has no standard member*. This is that sentence arriving for range, in a group whose own name — *putative Western Indo-Pacific* — carries a prediction about where its members ought to go. The animals did not read the name. *The variation is not error around a type; the type is a summary of the variation, and a poor one.*

**Three things to keep straight, one of them in the coverage** **n = 9, seven of them male, one archipelago, and tags that stop.** The authors say what that buys and what it does not: longer migrations could occur outside the tracking window, and the corresponding author’s own expectation, in the release, is that yellow-morph females *will* eventually leave to nest in the western Pacific, because they are not known to nest in the Galápagos at all. So “they stayed put” is true of 45 to 418 days per animal and is not yet true of a life. **The release says “the three years of our study” and the abstract says 45 to 418 days per turtle** — those are two different durations describing two different things, and only the second is the paper’s. **And the surprise in the headline is a quotation, not a result:** “We expected them to migrate to distant beaches in the western Pacific” is the corresponding author speaking in the release; the abstract says only that how these animals use local habitat “remains unknown.” Last, the arithmetic: **± 2,913.9 is labelled *SE* in the deposited abstract**, and a standard error nearly twice its own mean across nine animals is unusual enough that a standard deviation is the likelier reading — but we have not seen the methods, so we are quoting the label the paper deposited and declining to correct a statistic we cannot check.

**Where it goes** Galápagos, morphotype and site fidelity return zero across this site. [*Everywhere That Turtles Go*](https://starstuff.earth/everywhere-that-turtles-go-zine.html), No. 35, is about shells and breathing rather than range, so the migration argument is not spent by it; home range appears on [*A Field Guide to the Ways of Being Alone*](https://starstuff.earth/being-alone-field-guide.html), in a different sense. The natural home is a new entry in the turtle guide rather than a zine — *there is no standard distance from home* — and it owes the paper first, because an entry resting on a deposited abstract would be a field guide citing a deposit.

**Muñoz-Pérez, J. P., Alarcón-Ruales, D. E., Tanabe, L. K., Hart, C. E., Heidemeyer, M., Steiner, T., Arauz, R., Valle, C. A. & Rowe, J. W.** “Movements of yellow-morphotype green turtles (*Chelonia mydas*) in the Galápagos Marine Reserve.” *Frontiers in Amphibian and Reptile Science* **4**, online 27 August 2026. **DOI** 10.3389/famrs.2026.1837921. CC BY 4.0. **Read:** the publisher-deposited structured abstract at Crossref, in full; every figure and quotation attributed to the paper above is from it. The island-by-island movements, the nicknames, the April–October seasonality and the 2015 tagging date are from the Frontiers release and are marked as the release’s. **Not read:** the paper, which is CC BY with a published PDF address — www.frontiersin.org is refused at the network layer and no PubMed Central record exists. **Wire:** ScienceDaily, 27 September 2026.

Verified *Published* 2026-07-10 (issue) *Journal* Science Advances *Read* the paper, in full

Lag 77 d

### The children buried with adults were almost never their children

When a medieval grave holds more than one person, the reading has always been that they were family — not because anybody measured it, but because there was no way to. Children cannot be reliably sexed from their skeletons, early Christian burials carry no grave goods to read, and so children have largely been left out of the gendered archaeology of their own period. This team sequenced **142 new ancient genomes from three sites in Sweden** — Fjälkinge, Sigtuna and Västerhus, late Viking Age into the medieval period — **68 of them subadults**, and then ran kinship three separate ways, keeping only relationships all three methods agreed on. The result: at Västerhus, of 25 multiple burials with usable data from two or more people, **three — 12.0% — held first-degree relatives**, and *no* second-degree relationship was found within any single burial, though second-degree ties were common *across* burials. At Sigtuna the figure was **two of sixteen, 12.5%**. The abstract’s own sentence: “Contrary to the assumption that collective burials reflect close kinship, our results show that children interred with adults rarely shared close biological ties.” And the discussion’s: concepts of family and fellowship “extended well beyond the modern notion of the nuclear household.”

**The star stuff frame** **The assumption was never a finding. It was the default reading of a grave, held for as long as the alternative was unmeasurable** — and when the measurement arrived it came back at one in eight. This site already argues the present-tense version: [*The Village Was Never Just Family*](https://starstuff.earth/never-just-family-zine.html), No. 64, and [*A Field Guide to the Ways of Being Family*](https://starstuff.earth/being-family-field-guide.html), which is one of the guides whose cards are the families people make rather than the families people are issued. What is new here is the *period*: the nuclear household is being read backwards into the eleventh century by people who have no evidence for it and no way to get any, and now there is a percentage. The paper also notes at least one infant girl buried in a “typically male” context and more distantly related to the men beside her than to anyone else in the grave — one child, one grave, and exactly the kind of case a general rule would have eaten.

**Three sites, and a word the paper refuses to let do too much** **This is three cemeteries in one region in one period**, and 21 of the 115 newly sequenced individuals from multiple burials yielded too little data for kinship at all, which is why the denominators above are 25 and 16 rather than the full set. The authors are careful in a way worth copying: they define *kindred* as a purely genetic grouping that “does not imply any specific social, cultural, or household affiliations.” That cuts both ways, and the second way is the one to say out loud — genomics can exclude close biological kinship, and it cannot tell you what the relationship *was*. The absence of a father is not the presence of a fosterage. **And the paper reports a result against itself, which we are not going to quote without its other half:** the predominance of same-sex adult–child burials (17 of 28) survives a Monte Carlo simulation at *P* = 0.0002, and a Fisher’s exact test on the same contingency table finds no significant association, *P* = 0.4456. Both are in the abstract’s own results; the paper’s reading is that same-sex pairing was the most frequent practice without being dictated by the ages present. **The lag is a floor:** Crossref deposits no online-first date, only the 10 July issue.

**Where it goes** Mostly spent, and that is the right answer. The claim is the site’s own — chosen family appears on six pages, nuclear family on three. What is unspent is this piece of evidence: Viking appears twice here and in neither case about kinship, and no page carries an ancient-DNA kinship result of any kind. The honest placement is not a new zine; it is **a card in the family field guide and a sentence added to No. 64**, with the percentage and the caution attached — which is the smaller move and the one that does not re-argue a point the site has already made better.

**Krzewińska, M., Kjellström, A., Yaka, R., Rodríguez-Varela, R., Pochon, Z., Kempe Lagerholm, V., Hedenstierna-Jonson, C., Zachrisson, T., Kashuba, N., Sobrado, V., Naidoo, T., Vural, K. B., Jakobsson, M., Kılınç, G. M., Storå, J. & Götherström, A.** “Equal in death: Ancient genomic analysis of children’s early Christian burials.” *Science Advances* **12**(28), issue of 10 July 2026. **DOI** 10.1126/sciadv.aeb8588. **Read:** the published article in full at PubMed Central (PMC13353417), including methods and discussion; every figure and quotation above is from it. **Not read:** the supplementary sections and tables (S1–S10), which carry the pedigree reconstructions and the osteological detail. www.science.org returns 403. **Date note:** issue date only; the 77-day lag is a floor. **Wire:** ScienceDaily, 25 September 2026.

Verified *Published online* 2026-06-28 *Journal* Autism Research *Read* the paper, in full

Lag 88 d

### A rare-syndrome prevalence figure turns out to have been a measurement of who gets offered a genetic test

Phelan–McDermid syndrome is caused by haploinsufficiency of a single gene, *SHANK3*, through either a 22q13 deletion or a pathogenic sequence variant. Its published prevalence has been **2.5 to 10 per million births**. This team went to ten sources — clinical genetic testing laboratories, research centres and clinical centres — and asked each how many PMS diagnoses it had made out of its autism-tested participants: **179,837 autism cases** in total, with raw diagnosis frequencies running from 1% to about 2.5%, which the authors say must not be compared directly because the assays differ in what they can see. After adjusting for assay sensitivity, for people with PMS who are not autistic, and for the age at which autism is diagnosed, and then extrapolating through CDC autism rates, the weighted estimate is **13.7 per 100,000 (95% CI 10.02–18.60)**, about **1 in 7,300**. That is between fourteen and fifty-five times the figure it replaces. The paper’s explanation of the gap is not biological at all. It is a list of gates: whether a referral to genetics happens; whether testing is offered, when “a minority of individuals are offered testing” despite two decades of guidelines; whether insurance or Medicaid medical-necessity criteria — “often not aligned with current care guidelines” — will cover it; whether the test performed is a chromosomal microarray, which finds copy-number variants and so **misses roughly half of PMS**; whether the laboratory analyses *SHANK3* at all, since several do not, the gene being awkward; and finally whether a diagnosed family ever hears of the registry that counts them. Socio-economic status and distance from a tertiary centre are named outright.

**The star stuff frame** **The old number was never a fact about bodies. It was an instrument reading, and what the instrument was reading was access.** Everything that made the figure small sits in the environment and the systems around a person — a referral, a coverage criterion, an assay’s blind spot, a laboratory’s exclusion list, the distance to a hospital with a genetics department. Not one of those is a property of the people being counted, and all of them were doing the counting. The [rungs this house builds on](https://starstuff.earth/the-ladder-we-dont-print.html) are visible in that list from the bottom up, and the standing rule is that a piece puts the rungs in and leaves the ladder out. There is a second thing worth saying, and it is the harder one: *the early reports were of more severely affected people because only they were tested*, so the clinical picture of the syndrome was shaped by the same gate that shaped the count. **A category assembled from whoever got through the door will describe the door.**

**An extrapolation, and a frame we are not adopting** **This is a modelled estimate and the paper says so in its own title** — “Estimated To Be ~1:7300 Using a Multisource Model.” The denominator is autism-ascertained cohorts, so the component of the estimate covering people with PMS who are *not* autistic is itself an adjustment rather than a count, and the published interval, 10.02–18.60 per 100,000, is a sampling interval that does not carry the uncertainty in the adjustments stacked before it. The honest summary is a well-argued order-of-magnitude correction, not a census. **And the register is not ours.** The paper is written for funders and health systems — its stated uses are “disease burden,” “pharmacoeconomic modeling” and “resource allocation,” and it describes the syndrome in the clinical vocabulary of deficit throughout. That is a legitimate way to write a prevalence paper and it is not a frame this page can lift whole. *We are filing the finding and naming the frame, rather than quietly inheriting it.* The people the old figure missed were missed on the way to the test; that sentence is the paper’s own logic and it is the one we would carry.

**Where it goes** Phelan, SHANK3, ascertainment and underdiagnosed all return zero here. prevalence hits four pages, and the closest is [*Two Hundred Years Before You’re Grown*](https://starstuff.earth/two-hundred-years-zine.html), No. 34, which declines to print a population percentage for a parasite found on six sharks — the same caution, pointed at a much smaller claim. So the argument is unspent and the discipline for it is already in the house. The piece pairs with [the Enceladus item](#enceladus-1002) above: two measurements, one of ice and one of a population, both of which describe the thing that selected the sample. **It is not a Glimmers entry** — nothing happened on a particular day — and it is not a field-guide card. It is a [Star Stuff](https://starstuff.earth/collection-star-stuff.html) zine if it is anything, and it cannot be built from this paper alone: it needs the testing-access literature the paper cites and we have not opened.

**Levy, T., Lapidus, D., Friedman, K., Siper, P., Glass, L., Squires, L., Hames, M., Buxbaum, J. D. & Kolevzon, A.** “Prevalence of Phelan–McDermid Syndrome Estimated To Be ~1:7300 Using a Multisource Model.” *Autism Research* **19**(8), online 28 June 2026. **DOI** 10.1002/aur.70297. CC BY-NC-ND 4.0. **Read:** the paper in full at PubMed Central (PMC13471831), including the introduction’s account of the exclusion points; every figure and quotation above is from it. **Not read:** the supporting information, and the referral- and coverage-rate studies the introduction cites, which are the evidence for the gates and which this page has taken on the paper’s word. **Wire:** ScienceDaily, 24 September 2026.

Verified *Published online* 2026-06-30 *Journal* Integrative Organismal Biology *Read* the paper, in full

Lag 91 d

### The shark was described by an organ it does not have, and nobody could test it until a drone could hover at fifty metres

Sound in water travels through two regimes. Close to the source — within three wavelengths, the *near field* — it is a hydrodynamic sloshing of water particles. Further out, in the *far field*, it is a pressure wave, and a fish detects pressure by having a gas-filled swim bladder that squeezes and expands and re-radiates the signal to the inner ear as motion. Elasmobranchs have no swim bladder. From that absence the field inferred, reasonably and for decades, that a shark cannot detect sound past the near field at all — the paper’s own opening words are that sharks “are presumed to lack the capacity to detect these sounds beyond the acoustic near field.” This study flew a drone at 40–50 m over aggregating blacktip sharks off southeast Florida, played pulsed noise bands at 100–200, 200–400 and 400–800 Hz from an anchored underwater speaker, and used the known length of the boat to calibrate the distance at which each shark reacted. The sharks turned hard away — a mean change of trajectory of **72.7°**, to a mean heading of **136.1°** from the speaker — and **71.5% of the 165 responses were initiated in the acoustic far field**, from at least **62 metres**. They did not respond once to a 10 kHz control at the same volume, across 209 presentations. Lower frequencies needed less signal: about **16 dB** above ambient at 100–200 Hz, against about **45 dB** at 400–800 Hz. What the shark has instead of a swim bladder is the **macula neglecta**, a patch of sensory epithelium whose cupula carries no dense otoconia — and in carcharhinid sharks it is the largest, and the most hair-cell-rich, measured in any vertebrate.

**The star stuff frame** **An animal was characterised by a missing part, and the characterisation held because the experiment that would have tested it could not be built.** The chain was clean: no swim bladder, therefore no pressure transduction, therefore no far-field hearing. Each step is sound. The conclusion is about what the shark *cannot* do, derived entirely from what it does not have, and never once addressed to what it does have instead — an organ so developed in this family that it holds the vertebrate record. *This is the deficit model with a gill slit*, and the thing that broke it is not a new theory but a cheap drone: somebody could finally watch a wild shark and measure its distance from the sound at the same moment. The paper is quietly clear that the capability was not the obstacle — a 1964 conference abstract describes sharks orienting from 200-plus yards, observed from a circling aeroplane, and was never published. Sixty years on, the same observation with a measurement attached.

**The mechanism is not shown, and two of the paper’s own numbers disagree** The authors do not claim to have found the receptor: *“the mechanism remains unclear.”* They could not measure particle motion in the field, which is the quantity that actually matters, and they say so — “ideally, particle motion would have been measured as well, but this was not practical under the field conditions.” They also raise the objection against themselves, that sound energy bleeding below each nominal band would push the near field further out than modelled, and answer it with the 400–800 Hz responses, which cannot be explained that way. **Two internal figures do not match, and we are using the ones from the results.** The synopsis reads “71.6% of all responses (*n* = 209)”; the results report **165** responses (59 + 47 + 59) and attach *n* = 209 to the control presentations instead, computing **71.5%**. The discussion also gives the peak of the 100–200 Hz stimulus as 125 Hz where the results give 120 Hz. **And the coverage adds two things the paper does not.** “Nearly 250 feet” is about 76 m; the paper’s claim is *at least* 62 m (203 ft), and the farthest single response recorded was 74.0 m. “Pinpoint the source” is more than is shown: what the paper reports is directional information and a consistent turn away, and its own discussion floats the possibility that the consistency of that 72.7° is simply how far a blacktip’s body bends.

**Where it goes** **Half spent, and the spent half is the same shape.** [*A Field Guide to Being a Shark*](https://starstuff.earth/shark-field-guide.html) already carries the great hammerhead whose head, tested directly, *widens the search area* rather than sharpening the electrosense — an assumption about a body part overturned by measuring it. What is unspent is this organ and this inference: macula neglecta, swim bladder and far field each return zero across every page here. The guide is the place, not a zine — *there is no standard way to hear*, filed beside the hammerhead so the two cases sit together. If it ever becomes a zine, the argument is the inference and not the shark: **what you conclude from an absence depends entirely on whether you went looking for the substitute.**

**Sullivan, C. L., Gerstein, E. R. & Kajiura, S. M.** “Orientation of Blacktip Sharks (*Carcharhinus limbatus*) to Underwater Sound.” *Integrative Organismal Biology* **8**(1), obag033, online 30 June 2026. **DOI** 10.1093/iob/obag033. CC BY 4.0. **Read:** the paper in full at PubMed Central (PMC13398238), including methods, results, all four discussion sections and the acoustic-propagation measurements; every figure and quotation above is from it. **Not read:** Movie 1, the supporting video of a response. academic.oup.com is refused at the network layer. **Wire:** ScienceDaily, 29 September 2026.

## Held

Six this week, and they do not share a single pattern the way last week’s did. Two are held because the strongest number in the coverage is not the strongest number in the paper, and in one of those the figure the headline dropped is the one *we* would have led with. One is a review reported as a result. One is nine months older than the week it arrived in. And one is held for a reason this page has not had to write before: **the only summary of it available anywhere we could reach was written by a language model.** **Two of the six are papers we would have liked to file.** Nothing here was invented to fill the section, and nothing looked at and declined has been quietly omitted.

Held — the only available summary was generated by a model

### Scientists were wrong about this strange mammal for nearly 40 years

Zhelestids have been known since the 1980s, mostly from isolated teeth, and those teeth — low and rounded, built for plants where most Cretaceous mammals had sharp insect-eating cusps — led some palaeontologists to read them as early hoofed mammals, or as placentals living well before the end of the Cretaceous. A nearly complete skeleton from the eastern Gobi, found on a 2004 expedition and named *Tamirkhan balcarceli*, has the rounded zhelestid teeth and also long ever-growing incisors, a zalambdalestoid skull and long slender hind legs with distinctive ankles. The title is the conclusion: *Cretaceous zhelestid mammals are zalambdalestoids.* **This is a piece we would write** — a group read for forty years off the one part of it that fossilises best, with a coauthor quoted on Cuvier’s claim that a single tooth lets you predict a whole animal. **It is held because of what is reachable, and the specific shape of that is new.** Crossref deposits no abstract; www.nature.com is refused; no PubMed Central record exists. What an index *will* hand over is a tldr field containing fluent, paragraph-shaped prose about the paper, **generated by a language model and labelled as such in the same response**. It is accurate-sounding, specific, and not a source; nothing in its shape distinguishes it from an abstract, and a tool that did not read the label could not tell. We will not quote it and have not used it above — everything here is from the title and the American Museum of Natural History’s release, and is marked as the release’s. Worth chasing when the paper reaches an archive. **One check that did run, because it is the standing lesson:** this site names Cuvier twice already, and only one of the two is Georges — watching-animals-field-guide.html has *Cuvierian tubules*, a sea cucumber’s defence, and unfinished-animals-field-guide.html has the real one, citing Mary Anning while London displayed her specimens without her name. *Read the hits.* Giallombardo, A., Hoffman, E. A., O’Leary, M. A., Velazco, P. M., Zack, S. P. & Novacek, M. J., *Nature* **657**(8133), 973–980, online 29 July 2026, DOI 10.1038/s41586-026-10861-5, lag 58 days. **Read:** the title, authors, journal, pages and dates, from Crossref. **Not read:** everything else, including any abstract, because none exists in a form we are willing to treat as one.

Held — and the number the coverage dropped is the one we would have led with

### Brain scans reveal which disorders are linked to faster brain aging

Structural MRI from **45,900 controls and 2,698 patients** across nine conditions, scored by the predicted age difference — how much older a brain looks than its owner. The effects are large at one end: Alzheimer’s *d* = 0.97, combined alcohol-and-tobacco use disorder 0.84, tobacco use disorder 0.72, schizophrenia 0.53, major depressive disorder 0.28. **And at the other end they are nothing at all: autism *d* = 0.06 (95% CI −0.04 to 0.16, *p* = 0.36) and ADHD *d* = 0.01 (−0.14 to 0.15, *p* = 0.98)** — the paper’s own words are “not different from expected in developmental disorders.” The abstract’s results sentence nonetheless opens “while PAD was consistently greater across disorders” and reaches its two exceptions four clauses later, and the headline above keeps the opening and loses the exceptions. **So why is this held rather than filed?** Because filing it would mean accepting a predicted-age-difference as a measure of brain health in order to enjoy a null, and this page does not get to switch instruments depending on the answer. The study puts autism and ADHD in a list of disorders whose brains are being scored for abnormality, and the result we like is produced by that same scoring. *A null that flatters us is still a number from an instrument we do not accept.* We write from inside this; that is the reason for the care, not an exemption from it. The authors also name their own largest problem — the psychiatric and addiction groups have high comorbidity and “these potential confounders were not considered.” Liang, C., Pearlson, G., Bustillo, J., Kochunov, P., Chen, J. *et al.* (15 authors; Calhoun, V. D., last), *PLOS Medicine* **23**(7), e1004860, online 21 July 2026, DOI 10.1371/journal.pmed.1004860, CC BY 4.0, lag 68 days. **Read:** the publisher-deposited abstract at Crossref, in full; every figure above is from it. **Not read:** the paper, which is openly licensed and was not pursued to full text once the framing settled the hold.

Held — “twice as likely” is the weakest design’s number

### Cannabis users were twice as likely to commit violence, review finds

A PRISMA meta-analysis of **63 studies, N = 265,079**, searched to November 2024. The headline’s factor of two is in there: general-population odds ratio **2.05** (1.74–2.41), cross-sectional across designs **2.37** (1.66–3.40), psychiatric patients **2.49**. **And so is the number that is not in the headline: the longitudinal estimate is OR 1.17 (1.07–1.28).** Longitudinal designs are the only ones in this analysis that can speak to which thing came first; the association falls by most of its size when the design can address order in time, and the review reports both figures side by side without resolving the gap. A finding that shrinks that much under a better design is telling you something about its own literature, and the coverage took the larger number. Held for that, and held again for the register, which is the one this site exists against: a population-level association between a thing people do and a thing people fear, addressed to a reader as though it described them. The paper itself is more careful than its headline — it frames the question through psychosis and names victimisation as well as perpetration (OR 1.49), which is the half nobody reported. Trotta, G., Rodriguez, V., Marino, P., Gurmesa, M., Spinazzola, E., Li, Z., Alameda, L., Di Forti, M., Murray, R. & Vassos, E., *Psychological Medicine* **56**, online 24 September 2026, DOI 10.1017/S0033291726105856, lag 6 days. **Read:** the publisher-deposited abstract at Crossref, in full; every figure above is from it. **Not read:** the paper.

Held — it is a review, reported as a result

### Mulberry may reshape gut bacteria and influence metabolism

There is no cohort here, no trial and no new measurement: the paper is a narrative review of what mulberry leaf and fruit preparations have done in *preclinical* models of metabolic dysfunction, and its interest is really in processing — that drying, maceration, fermentation and extraction change the chemistry enough that two mulberry products are not the same object. Its own verbs say what it is. Available preclinical studies “indicate”; reported findings “frequently include” enrichment of taxa “commonly regarded as beneficial”; technological factors “may substantially influence” the result. **“May reshape gut bacteria and influence metabolism” is a fair description of a review’s subject matter and a poor description of anything that happened to anyone.** Held on that, and the house would hold it regardless: a plant preparation, a list of bacteria pre-labelled beneficial, and a reader who will read it as advice. Miszczak, M. M., Kłosowska-Buryło, K., Pieczyńska, J. M., Bielecka, M. & Prescha, A., *Biomolecules* **16**(7), 965, online 30 June 2026, DOI 10.3390/biom16070965, lag 90 days. **Read:** the publisher-deposited abstract at Crossref, in full; the quoted verbs are verbatim from it. **Not read:** the paper.

Held — lag 197 days, the longest thing on this page, and “2.5 years” is an equivalence

### This diet may slow brain aging by 2.5 years

**1,647 people from the Framingham Heart Study Offspring cohort**, MIND diet scores from a food-frequency questionnaire administered at three exams, brain imaging between 1999 and 2019 with a median of three scans each, median follow-up **12.3 years**. Greater adherence was associated with slower loss of total grey matter: each three-unit increase in the score with a **0.279 cm&sup3;/year** (95% CI 0.089–0.469) slower decline, which the authors describe as “a 20.1% attenuation in age-related change that was equivalent to 2.5 years of reduced brain ageing during the 12.3-year follow-up.” **That is an equivalence computed from a volume slope, not a measurement of brain age and not 2.5 years of anything** — and it is an association in an observational cohort, where what people eat is entangled with everything else about how they live. Held firstly for that, and firstly also for the date: the paper went online on **17 March 2026** and the wire carried it on 30 September, a lag of **197 days**, longer than every filed item on this page and longer than the 128-day hold below. A digest whose axis is *how long did that take to reach you* cannot run a six-and-a-half-month-old cohort result under a present-tense headline without saying so. Chen, H., Hailili, G., Tong, L., Fei, L., Cao, Y., Xu, X., Li, X., Melo van Lent, D. & Yuan, C., *Journal of Neurology, Neurosurgery & Psychiatry* **97**(6), 505–512, online 17 March 2026, DOI 10.1136/jnnp-2025-336957, lag 197 days. **Read:** the publisher-deposited abstract at Crossref, in full; the quotation and both figures are verbatim from it. **Not read:** the paper.

Held — the abstract we have came from a third-party index, and the headline’s claim is the paper’s inference

### 1.7-billion-year-old fossils reveal a crucial clue to the rise of complex life

An integrated palaeontological, sedimentological and geochemical reconstruction of where the oldest known eukaryote fossils, roughly **1.75 to 1.4 billion years old**, actually lived. The answer is a distribution: the fossils turn up across environments from coastal to offshore, but “almost entirely restricted to those from settings with oxygenated bottom waters,” and their near-absence from otherwise fossil-bearing anoxic samples suggests they were sitting on the seafloor rather than drifting, since plankton would show up in both. The proposal is that eukaryotes were confined to oxic benthic habitats for most of the Proterozoic and only reached open water in the Neoproterozoic — which, if it holds, explains a mismatch that has bothered this field for years, between when eukaryote body fossils appear and when their molecular biomarkers do. **Held on two counts and we would like to read it.** First, the abstract above **came from Semantic Scholar’s index, not from the publisher** — Crossref deposits none for this paper and www.nature.com is refused — and an index’s copy of an abstract is one remove further out than this page’s plausible grade allows, even when the words are the authors’. Second, the sentence the coverage leans on is softer in the abstract than in the headline: the mitochondria claim is explicitly an inference, “given their size and morphological complexity, probably possessed mitochondria,” drawn from where fossils occur rather than from anything seen inside a cell. Lag **128 days**, the second-longest on the page. Lechte, M. A., Riedman, L. A., Porter, S. M., Halverson, G. P. & Whelan, M., *Nature* **655**(8123), 670–675, online 20 May 2026, DOI 10.1038/s41586-026-10533-4, lag 128 days. **Read:** the abstract as held by Semantic Scholar, in full, and the bibliographic record at Crossref. **Not read:** the paper, and no publisher-deposited abstract exists to read.

## Seedbed

Three seeds, two of them held over and now carrying a second and third case — which is what a seedbed is supposed to do and what this one has not previously managed. Plus the leads nobody opened, and a ruling on the lead list, which has now grown for three weeks running without anything coming off it.

Notes page · third case, and the first outside physics

### Three ways of arranging not to be able to steer the answer

This seed has been waiting for the thing it got this week. [On 18 September](https://starstuff.earth/glimmer-wire-2026-09-18.html#seed-consensus-0918) it started as a survey in which physicists published the distribution of their own belief rather than a spokesperson’s version of it, and several positions described publicly as consensus turned out to be narrow majorities. [On 25 September](https://starstuff.earth/glimmer-wire-2026-09-25.html#seed-envelope-0925) it acquired a laboratory that scrambled a decade of gravity data against a key sealed in an envelope, so that nobody could stop adjusting at the moment the number looked right. **Both were physics, and the note said so: two pictures from one discipline is a portrait of physics, not an argument about method.** The [model scroll](#scroll-1002) filed above is the third case and it is archaeology and materials science, and it is the clearest of the three, because the discipline is the object: they made a scroll whose text they already knew, so that a reading algorithm can be marked rather than admired. **What the three have in common is now sayable, and it is not “they were careful.”** In each one, the people producing the result *arranged in advance to be unable to influence it* — a distribution in place of a spokesperson, a key in an envelope, a text known before the instrument saw it. That is a shape, and it might be the page. *What is still missing:* the *Metrologia* paper, which is CC BY and still unreachable and is the only document that says what was blinded and how; the survey preprint, still unread and still without a version of record; a fourth case from a field where the stakes are human rather than physical, because all three of these are about objects that cannot be harmed by being got wrong; and a title, which the page does not have and will not get from a list.

From the *PLOS One* paper filed above, read in full, plus the two earlier seeds on their own pages. **Not obtained:** Schlamminger *et al.* in *Metrologia*; arXiv:2605.11058; any case outside the physical sciences.

Zine · How We Got Here, and now it has two cases

### A name moves in four directions in one paper, and the animal does not move at all

[Last week’s strongest seed](https://starstuff.earth/glimmer-wire-2026-09-25.html#seed-drawer-0925) was a penguin described as a new species in 2026 from a skin collected on 31 December 1897, in a paper that simultaneously demoted to subspecies a population another team had raised to species rank in 2020. One paper, two directions. **Seven days later the [Anchiale revision](#anchiale-1002) does four at once**, in a different phylum and an unrelated literature: two species created, one resurrected from inside another name after a century and a quarter, one name sunk as a synonym along with the replacement name coined for it in 2007, and three names permanently fixed to chosen specimens by the designation of lectotypes. **Two independent cases a week apart is the difference between an anecdote and a form**, and the form is the piece: a chain through how a name gets made, unmade, revived and pinned, with every joint marked, because both papers document their own acts in the formal vocabulary of a code designed for exactly this. The reader it is for is the one who has been handed a diagnosis, a label or a category as a discovery about themselves. The argument is not that the categories are arbitrary — both of these are evidence-led and better than what they replaced — but that *they have dates, authors and a procedure for changing, and the organism is unaffected by every revision.* *What is missing:* both papers in full, since Wiley and the gentoo supplementary tables are equally out of reach; Tyler *et al.* (2020) on *P. poncetii*, which a zine about taxonomic rank cannot cite at one remove; a checklist authority’s current treatment of either group, which is the real test of whether a proposal has become the answer. **And the drawing problem from last week has got worse rather than better:** four gentoos differing mostly in size, plus five stick insects differing in egg lids and the shape of a thoracic spine, is nine animals that will read on paper as two animals drawn badly. The honest figure is not the organisms. It is the nomenclature itself — a timeline of names, branching and merging and being nailed down, with the unchanged animal along the bottom.

From the *Austral Entomology* abstract filed above and the *Communications Biology* paper read in full last week. **Not obtained:** Jones *et al.* in full; the gentoo Supplementary Tables 9 and 13; Tyler *et al.* (2020); any checklist authority’s current treatment.

Field guide entry · off a paper read in full

### The weapon was already inside the nutrient

Lactoferrin is the iron-binding protein in milk and tears, and everybody has known what it is for since the 1960s. Inside it, as a stretch of its own sequence rather than a protein in its own right, sits **lactoferricin** — a cryptic antimicrobial peptide that ruptures bacterial membranes. This paper resurrects extinct lactoferrin ancestors back to the earliest mammals and watches that embedded stretch gather cationic and hydrophobic residues over time, until the ancient version could first break a membrane and later versions could do it well; and it pins a single rapidly evolving site among great apes that measurably raises potency against major pathogens. **The entry writes itself around what a cryptic domain is:** a function that was never announced, inside a molecule already fully named for something else, arriving by degrees rather than by appearing. There is no moment at which lactoferricin starts existing. *What holds it at seed:* ancestral sequence reconstruction is a model of a protein nobody has ever met, and reading how much confidence it deserves *is* the work — we have the paper and not the alignments or the posterior probabilities behind each resurrected residue. And the coverage’s phrase, “160-million-year-old proteins,” is not the paper’s object: what was tested in the laboratory is a reconstruction dated by the mammalian tree, not a protein anybody dug up. The guide it would join needs choosing carefully, too: the twelve guides that arrive at *there is no standard* something are a set this site has already caught itself miscounting by grep, so the set has to be re-derived and the hits read before anything is added beside them.

Sil, T., Kowalski, C. H., Scamfer, S., Copeland, N. & Barber, M. F., “Retracing the origin and evolution of a cryptic antimicrobial peptide within mammalian lactoferrin,” *PLOS Biology* **24**(8), e3003932, online 25 August 2026, DOI 10.1371/journal.pbio.3003932, CC BY 4.0 — read in full at journals.plos.org; wire ScienceDaily, 25 September 2026, lag 31 days. **Not obtained:** the supporting alignments, the ancestral-reconstruction confidence values, and the structural work behind the great-ape site.

Leads · unchecked, and a ruling on the list

### Three off this week’s wires, and the last day of the growing list

This week’s unopened: a *Geology* paper arguing that a two-billion-year-old carbon signature in the Zaonega Formation is thermogenic rather than what it has been read as, which would be a correction of a long-standing clue about early Earth and which we have as a title and a bibliographic record only. A *Nature* paper deriving the multidecadal wobble in the length of Earth’s day from gravitational torque in the deep interior — no abstract deposited anywhere, and www.nature.com refused. And a JWST/MIRI result on protoplanetary disk winds evolving from molecular to atomic, which is a development-and-timing story of the kind this page favours, is CC BY, and has a 4,588-character abstract sitting deposited at Crossref that nobody here opened past its title. Two more exist as headlines and nothing else: a 245-million-year-old fossil preserving its stomach and intestines, and a broken rib in the largest known *T. rex*. **None of these was read.** **And now the part that matters.** Last week’s four leads — the aerodynamically sorted planetesimals, the magnetotactic-bacteria nematodes, the first extragalactic stellar stream, the undated crater — were not chased. The week before’s four were not chased either. The monotile Supplementary Information and the two Smith *et al.* papers that edition four’s best seed depends on are still unread. That is **three consecutive weeks of a list that only grows**, and last week this page wrote that such a list is “a to-do list pretending to be a method.” Writing it twice does not make it a method either. *So: next week this list is either worked or it is closed.* If a lead is still unopened a fortnight after it is written down, it was never a lead; it was a note about how interesting the week looked.

Thiagarajan, N., Lepland, A., Eichinger, F., Prave, A. & Eiler, J., *Geology* **54**(10), 1015–1020, DOI 10.1130/G54338.1, online 13 July 2026 — abstract deposited at Crossref, unread beyond its title. Zhang, H. & Dumberry, M., *Nature* **658**(8134), 119–124, DOI 10.1038/s41586-026-10999-2, online 23 September 2026 — no abstract deposited. Bajaj, N. S., Pascucci, I., Cabrit, S. *et al.*, *The Astronomical Journal* **172**(3), 161, DOI 10.3847/1538-3881/ae9089, online 25 August 2026, CC BY — abstract deposited, unread. The fossil and the rib: ScienceDaily listings of 27 and 23 September 2026, with no journal reference obtained for either.

## Not

**Not a citation.** Six papers were read end to end this week and the standing rule is untouched. Nothing filed here may be promoted to a zine, a [Glimmers](https://starstuff.earth/collection-glimmers.html) entry or a card without the primary source in hand, a [changelog](https://starstuff.earth/changelog.html) entry and its row in the fact-check ledger. For the three items resting on a publisher’s deposited abstract, lifting a figure out and printing it elsewhere would be laundering a deposit into a fact. **And one refusal is new this week:** a fluent summary returned by an index in the field where an abstract goes, generated by a language model and labelled as such, is not a source at any grade. We did not quote it, and the paper it belonged to is held rather than filed.

**Not “science says you are valid.”** Two items make that offer this week and the second is the sharper one. A measure of brain ageing finds no difference at all in autism and ADHD, and it would be very easy to carry that number around. *We are not going to* — because the instrument that produced the null is the same instrument scoring those brains for abnormality in the first place, and a page cannot accept a method only when it agrees. The belonging claim was never waiting on an effect size, and it does not move when the next study finds one.

**Not a deficit.** A prevalence figure for a genetic syndrome rose by between fourteen and fifty-five times, and what changed was the counting: referrals, insurance criteria, an assay that cannot see half the cases, laboratories that skip an awkward gene. *Nobody became more common.* We write from inside this, and the people the old number missed were missed on the way to the test. The paper’s own vocabulary — burden, allocation, modelling — is named on this page rather than borrowed from it.

**Not a superpower either.** A shark hearing from 62 metres through an organ nobody credited, nine turtles whose home ranges disagree with each other by twice their own average, five stick insects where there were two: these are facts about *architecture, history and range*, never about merit. The paper on the sharks does not say blacktips hear better than you; it says the inference drawn from their missing swim bladder was wrong. Flipping a deficit to a gift keeps the ranking and changes the sign.

**Not a cure list.** A diet associated with a slower grey-matter slope in an observational cohort is not 2.5 years of anything a reader can buy, and a review of mulberry preparations in animal models of metabolic dysfunction is not a thing to do on Tuesday. Both are held above, and held with their numbers printed, because the figures are real and the sentence built on them is not.

**Not anti-science.** Three corrections on this page are of positions the field itself held for decades — that a shark without a swim bladder cannot hear into the far field, that a group of Cretaceous mammals could be placed from its teeth, that people sharing a grave shared a parent. Every one was overturned from the inside, by the same discipline, with instruments it built on purpose: a drone that could hover and a boat of known length, a nearly complete skeleton, three kinship algorithms made to disagree with each other. *None of those corrections is an argument against evidence; each is an argument for the part that was missing.* If a draft off this digest starts reading as a case against the field, it has gone wrong.

**Not *because of*.** An ocean under the ice of a small moon of Saturn is sorting its own salts as it freezes, and the grains carry the record of the sorting rather than of the sea. That 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.

[Stimpunks Foundation](https://stimpunks.org/) × [More Realms](https://morerealms.com/) · Glimmer Wire · L★S
 Edition Six of the weekly scan · published 2 October 2026
 Every edition is kept on its own page · nothing deleted, the Held list included
 A working scan, not a finding · off the reading chain, on purpose
 [All editions](https://starstuff.earth/collection-glimmer-wire.html) · [Glimmers](https://starstuff.earth/collection-glimmers.html) · [Changelog](https://starstuff.earth/changelog.html) · [Too Good to Check](https://starstuff.earth/too-good-to-check.html)
 stimpunks.org · morerealms.com · starstuff.earth · [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/)