Sorted by lag, shortest first. Bars are scaled to the longest item, 157 days; the two-day item would draw a bar too thin to see and is given a visible floor of 2% rather than a true-to-scale sliver, which is a drawing decision and not a measurement. Three of the seven reached us within a month of publication — lags of 2, 15 and 31 days. The other four had been sitting in their journals for 85 to 157 days before a feed handed them over this week as new.
Plausible
Published 2026-09-17 (issue)
Journal Science
Read the authors’ preprint, in full
Lag 2 d
The belief that difference destabilises a system turns out to be an artefact of the shortcut that made the maths possible
For decades the working understanding in network dynamics has been that heterogeneity among nodes inhibits stability — that a network of oscillators synchronises best when its members are nearly identical. The authors’ first move is to point out that this sits awkwardly beside the world: the natural and engineered networks that actually hold together are heterogeneous almost without exception. Their result is that the received conclusion “arises from model reductions introduced for mathematical tractability and breaks down when nodal dynamics are higher-dimensional.” Collapse each node to a single phase variable — the Kuramoto reduction, with the master stability function and its relatives derived, in the paper’s words, “on the premise of oscillator homogeneity” — and you get a tractable model whose Jacobian is well behaved and in which sameness wins. Let each node keep more than one dimension, so the Jacobian is non-Hermitian, and the sign flips: heterogeneity enhances stability, “even when parameters are randomly disordered.” They show it for neural, power-grid and material networks, and show that in one-dimensional dynamics the same stabilisation can be bought instead through nonreciprocal interactions — an interaction that is not the same in both directions. The abstract’s closing sentence is the claim: disorder is revealed “not as a liability but as a general resource for stabilizing complex systems.”
The star stuff frame
The interesting part is not the conclusion, it is the
shape of the error. Nobody chose to believe that uniformity is what holds a system together. Somebody chose a simplification, because the honest problem was too hard to compute; the simplification was published, taught, and built into the standard tool; and a generation later the assumption inside it was being read back out as a finding about nature. The paper says so in its own words — the models were reduced
for tractability, and reduced models “may misrepresent the impact of parameter choices and network structures.” That is the same shape as
Five Sigma, No. 40, where a threshold chosen in 1925 with the word
convenient hardened into a fact about evidence — and it is arguably the more dangerous version, because a threshold announces that somebody picked it and a model reduction does not.
What this is about, and the grade
This is a result about coupled dynamical systems, and it is not a result about people. The nodes are oscillators, buses on a grid and species in a Lotka–Volterra community; the stability is linear stability at an equilibrium; the disorder is dispersion in nodal parameters. Nothing in it licenses a sentence about human variation, and the closest the paper comes to biology is a class of ecological models, where the authors say disordered networks offer “a natural route toward resolving May’s longstanding complexity-stability paradox.” That is a claim about a longstanding puzzle in theoretical ecology — the authors’ word is longstanding and we have not dated it — and not about an ecosystem anybody has measured. Graded plausible, not verified, and the reason is the rule and not the reading. science.org answers with 403 and doi.org is refused, so what was read is the authors’ arXiv posting — their text, complete, including the propositions and proofs, and still not the version of record. This page wrote that rule down on 11 September and is now applying it for the second week running to a paper it would have liked to file higher.
Where it goes
Checked rather than assumed, against the site as it stood
before this page existed — every count below excludes this scan, which now contains all of these words. The honest answer is
half spent. The conclusion is already the site’s:
The Universe Runs on Difference, No. 8, argues that variety is resilience and monoculture the fragile engineered thing, and
monoculture appears on 41 pages here. What is unspent is the
mechanism of the mistake: a simplification adopted so the equations would close, mistaken afterwards for a property of the world. No page here contains
Kuramoto,
complexity-stability or
power grid, checked by grep. That is a
How We Got Here chain rather than a Star Stuff zine, and its first joint —
the reduction was introduced for tractability — would be marked
documented, because the authors document it.
Montanari, A. N., Zanin, P. & Motter, A. E. “Disorder-promoted stability.” Science 393(6817), 1241–1249, issue of 17 September 2026. DOI 10.1126/science.aeg3946. Read: arXiv:2609.25226v1, 39 pages, submitted 21 September 2026 — the authors’ own text, read in full, and the source of every quotation above. Not read: the version of record; science.org returns 403. Date note: Crossref deposits no online-first date for this paper, only the print date of 17 September, so the 2-day lag is computed from an issue date. Preprint note: the arXiv posting is four days later than the journal, so unlike most weeks the preprint does not lengthen the real interval. Wire: ScienceDaily, 19 September 2026.
Plausible
Published online 2026-09-09
Journal JAMA Psychiatry
Read the abstract, deposited at Crossref
Lag 15 d
The people being diagnosed now look more like everybody else — which is a fact about the boundary, not about us
A matched case-control study built on the Danish national registers: 2,194,951 children and adolescents born between 1994 and 2022, of whom 71,317 were diagnosed with ADHD or autism before eighteen between 2012 and 2022, each matched to ten controls on sex, birth year, municipality and country of birth — 713,170 of them. Nineteen prediagnostic characteristics were tested: parental and family factors, birth and perinatal factors, health care use. Diagnosed young people differed from controls on all nineteen, and the differences shrank, year by year, across the decade. Low birth weight ran at an odds ratio of 1.54 (95% CI 1.41–1.68) in 2012–13 and 1.17 (1.10–1.24) in 2020–22. The attenuation is much stronger for ADHD than for autism — household income moves 6.5% a year (95% CI 5.7–7.3) for ADHD against 0.7% (−0.1 to 1.5) for autism, an interval that includes zero — and stronger for those diagnosed at 10 to 17 than for younger children.
The star stuff frame
The finding, stated as the paper states it, is that the boundary moved. The characteristics that used to mark out the diagnosed population are marking it out less and less, which is what happens when a threshold is drawn in a different place, not what happens when a population changes. We write from inside this, and the version of the story that reaches us is usually overdiagnosis — an accusation that some of us are not really what we say we are. The paper closes off that reading itself, in its own last sentence: the results “should not be interpreted as evidence that ADHD or ASD have become less impairing at the individual level.” That clause is the whole item. A person newly inside the boundary is not a person who was fine all along; they are a person who was outside it, and the old line is what needed explaining.
The words the paper uses, and the ones it is careful about
Its vocabulary is not ours and is worth naming rather than quietly translating. The nineteen characteristics are called “well-defined risk factors,” and the outcome is a diagnosis; a study designed this way cannot say anything about the people it counts except that they were counted. The causal claim is carefully hedged and the hedge is the paper’s own: increasing rates “may reflect changes in diagnostic practices or health care capacity rather than changes in underlying risk alone” — alone being a word that survives into no headline anywhere. This is one country with one health system and universal registers, which is exactly why the data exist and exactly why the result may not travel. And the ADHD/autism asymmetry is a real number with a real interval: the autism figure of 0.7% a year has a confidence interval running through zero, so “autism attenuated more slowly” is supported and “autism attenuated” on its own is not.
Where it goes
Checked rather than assumed, against the site as it stood before this page existed.
overdiagnosis and
diagnostic expansion appear on no page here, by grep;
diagnostic criteria appears on ten. The unspent argument is not about prevalence at all — it is that a diagnostic boundary is a
decision that can be watched moving in registry data, in a way a trait cannot. That belongs beside
The Name Was Already the Verdict, No. 88, where a category did the work and the evidence underneath it was assembled afterwards.
Tarp, M. E., Lousdal, M. L., Rask, C. U., Keyes, K. M., Formánek, T., Brikell, I., Liu, X., Maseras, G. T. & Plana-Ripoll, O. “Changes in Characteristics Associated With ADHD and ASD Diagnoses Over Time.” JAMA Psychiatry, published online 9 September 2026. DOI 10.1001/jamapsychiatry.2026.2722. Population-based matched case-control study, Danish nationwide registers, conditional logistic regression; data analysed April 2025 to June 2026. Read: the publisher-deposited structured abstract at Crossref, in full, and every figure above is from it. Not read: the paper. No PubMed Central record exists for it. Not the paper’s own: the lead institutions named by the release (ISGlobal and Aarhus) are the release’s attribution, not a figure we checked. Wire: ScienceDaily, 24 September 2026.
Verified
Published online 2026-08-24
Journal PNAS
Read the paper, in full
Lag 31 d
The oldest solids in the solar system are still holding a magnetic field, and it is stronger than the one you are standing in
Calcium-aluminium-rich inclusions — CAIs, millimetre-scale white flecks in chondrite meteorites, the oldest known solar system solids — carry a remanent magnetisation, and the authors measure what field they cooled in. The answer is roughly 150 to 600 µT, against the few tens of microtesla at the Earth’s surface today. The question that number settles is which force was moving matter in the first act of the solar system: the paper opens by saying it “is unknown whether magnetism and/or gravity dominantly drove accretion in the youngest evolutionary stages” of protoplanetary disks, and closes by reporting an intensity “consistent with magnetic fields playing a key role driving disk accretion while also heating the very inner disk to 103 K at the earliest stages of solar system formation.” More than 90% of a star’s mass is accreted in the embedded phase these grains date from, and the disk at that stage is invisible: wrapped in an infalling envelope, unobservable by telescope. So the only surviving instrument pointed at it is the rock itself.
The star stuff frame
This is the founding claim of the whole collection, arriving as an instrument reading rather than as a metaphor. The atoms were made in stars; these particular flecks are the first things that were ever solid after that, and they have been carrying a record of the conditions they condensed in for four and a half billion years without being asked to. A magnetic field is not a thing that leaves fossils — and it left one, in the only material old enough to have been there. What the grains preserve is not their composition but the state of the room, which is a different and stranger kind of inheritance.
The error bar is two-thirds of the number, and the paper prints it
The headline range comes out of a chain of corrections, and the paper’s own corrected figure is 324.2 ± 218.3 µT — an uncertainty of about 67% of the value, wider than most of the intervals this page has flagged. The lower bound of that interval still clears the Earth’s present field, which is what makes the qualitative claim survive the quantitative wobble, and it is the kind of distinction that vanishes in one retelling. Two conditionals in the paper are load-bearing and are the authors’: “If we assume that the cooling rates measured for CAIs apply to the CAIs studied here,” the measurements are a near-instantaneous record of the field; and the age of that record is the age of the last event that heated the grains above the iron-nickel magnetic blocking temperature, which need not be their formation. The result is also “consistent with” magnetism driving accretion — support for a hypothesis, not the elimination of gravity.
Where it goes
Checked rather than assumed, against the site as it stood before this page existed. supernova appears on 30 pages and nebula on 11, but calcium-aluminum, protoplanetary, solar nebula and presolar appear on none of them. The site has said repeatedly where the atoms were forged and has never once described the room they arrived into — a disk too young to see, too hot to survive, with a field strong enough to move the mass around. That is a Star Stuff register piece with one settled measurement at its centre, and it wants the wide error bar on its face rather than in its colophon.
Borlina, C. S., Weiss, B. P., Bai, X.-N., Tung, P.-Y., Harrison, R. J., Mansbach, E. N., Chatterjee, N., Tissot, F. L. H. & McKeegan, K. D. “Paleomagnetic evidence for a nebular magnetic field from calcium-aluminum-rich inclusions.” Proceedings of the National Academy of Sciences 123(36), online 24 August 2026, in the issue of 8 September. DOI 10.1073/pnas.2521660123. CC BY-NC-ND 4.0. Read: the paper in full at PubMed Central, PMC13552906, including the paleointensity corrections and the discussion. Not read: the supporting information, so the anhysteretic-remanence calibration behind the 324.2 ± 218.3 figure is reported as the paper states it rather than checked. Wire: ScienceDaily, 24 September 2026.
Verified
Published online 2026-06-30
Journal Nature Communications
Read the paper, in full
Lag 85 d
Blocking the enzyme did nothing. Removing the protein did everything. Those were never the same experiment
6-thioguanine has been given to people since the 1950s, for leukaemia, autoimmune conditions and inflammatory bowel disease. A genome-wide screen had flagged an enzyme called NUDT5 as a mediator of its toxicity, and knocking the gene out makes cells resistant — the authors reproduce that, a four-fold increase in the concentration needed to kill half the cells. The obvious inference is that NUDT5’s catalytic activity is doing the work. So they tested it with two chemically distinct inhibitors, and report the result in the flattest possible words: “Contrary to expectations, neither inhibitor was able to phenocopy the NUDT5 KO.” They then built degraders — molecules that destroy the protein instead of blocking it — and those did reproduce the knockout. The confirming experiment is the one that settles it: they put a catalytically dead mutant back into the knockout cells, and sensitivity to the drug came back. The protein does not need to work. It needs to be there. Their generalisation is a sentence about method rather than about this molecule: “for some enzymes, phenotypes attributed to catalytic activity may instead reflect protein-dependent functions that are selectively interrogated by targeted degradation.”
The star stuff frame
An enzyme is defined by what it catalyses, so for as long as the only available tool was an inhibitor, the only available question was what does its activity do? — and an unasked question comes back looking exactly like a settled one. The correction here is not that anybody reasoned badly; it is that a new instrument made a second question askable, and the second question had a different answer. This site has a standing line about a finding’s absence being a fact about where attention went. This is the sharper version: the absence was a fact about what could be done in a dish, and no amount of care would have found it with the older tool.
The headline is about a drug; the paper is about a protein
“Scientists thought they knew how this 70-year-old leukemia drug worked” overstates the subject, and the subject is the whole thing. 6-thioguanine’s own mechanism — a prodrug metabolised and incorporated into DNA and RNA — is not what changed; the paper describes it that way itself. What changed is the account of one modifier of the cellular response. That is a real correction and a narrower one than a reader of the headline would carry away. Everything here is in cell lines: HAP1, derived from a chronic myelogenous leukaemia line, and HL-60. The clinical sentence in the abstract is properly conditional — depletion of NUDT5 is antagonistic to NUDT15 inhibition, “suggesting a distinct mode-of-action with potential implications for patient therapy” — and no patient has been treated on the strength of it. One of the two probes used, MRK-952, is listed in the paper as to be published, so its selectivity is not independently checkable from here.
Where it goes
Checked rather than assumed, against the site as it stood before this page existed. mechanism of action appears on no page here and enzyme on eleven. The transferable line — switching a thing off and taking it away are two different experiments, and for most of the history of a field only one of them was possible — is unspent, and it generalises past biochemistry in a direction this site cares about, because a great deal of what gets said about accommodations is an inhibitor experiment reported as a knockout. That is a spread rather than a whole piece, and it would need the biology kept literal, because the moment it becomes a metaphor it stops being evidence.
Marques, A.-S. M. C., Bauer, L. G., Nguyen, T.-A., Gonzalez Orta, A., Venne, J.-L., Cheng, C., Balıkçı, E., Liu, Y., Tivon, B., Kroupova, A., Ciulli, A., London, N., Kubicek, S. & Huber, K. V. M. “Targeted Protein Degradation of NUDT5 Dissociates Catalytic Inhibition from Protein Loss in 6-Thioguanine Response.” Nature Communications 17(1), online 30 June 2026. DOI 10.1038/s41467-026-74489-9. CC BY 4.0. Read: the paper in full at PubMed Central, PMC13462928, including the figure legends and the discussion; both quotations are verbatim from it. Not read: the supplementary figures, so the metabolomic result on IMP is reported as the paper states it; and the second, much larger paper the same release cites alongside this one (Nguyen et al., 40-plus authors) was not opened at all, and nothing above rests on it. Wire: ScienceDaily, 23 September 2026.
Verified
Published online 2026-05-05
Journal Nature Communications
Read the paper, in full
Lag 140 d
One kind of tree, three different ways of breathing, growing side by side
In 1800 Alexander von Humboldt held a leaf of a tropical tree under water and noticed it doing something odd. The something was crassulacean acid metabolism: opening the stomata at night, fixing carbon dioxide into an acid in the dark, and releasing it internally by day with the pores shut — photosynthesis for a plant that cannot afford to lose the water. This paper sequences and compares three Clusia genomes chosen to span the range: C. major (weak CAM), C. minor (inducible, switching as conditions demand) and C. rosea (strong CAM). The word the field uses for these is physiotypes, and the genus is famous for holding them all — the reference the paper leans on, by one of its own co-authors, is titled “One morphotype, three physiotypes: sympatric species of Clusia with obligate C3 photosynthesis, obligate CAM and C3-CAM intermediate behaviour” (Lüttge, 1999). Sympatric means growing in the same place. One morphotype means you could not tell them apart by looking. The genomic finding is that all three are diploidised genomes of polyploid origin, and that whole-genome duplication followed by transposon-mediated diploidisation may have shaped the CAM gene families — particularly the ones governing how leaf starch is broken down to supply the acceptor molecule the night chemistry runs on.
The star stuff frame
Twelve field guides on this site arrive at there is no standard star, nest, egg, shark, way to rest. Here is the same sentence about the most basic thing a plant does. These trees are the same shape, in the same soil, in the same light, and they are running three different metabolisms — and the intermediate one is not a halfway house or a failure to commit, it is a plant that keeps both options and chooses by conditions. The literature the paper cites calls CAM a continuum rather than a category. It is unusually hard to read that as deficit, because there is no baseline: the question which of these three is the normal Clusia has no answer, and 1999 is when somebody wrote that down.
Three genomes, and a mechanism the paper marks as speculation
The causal claim is hedged twice by the authors and once by the journal. Polyploidisation and diploidisation “could have shaped” the CAM gene families; whole-genome duplication “might have driven” the diversification; and the editors’ own summary line says the authors “speculate” about it. That is three genomes from a genus of several hundred species, a comparison rather than a test, and it establishes a genomic framework rather than a history. What is not hedged is the phenotype. The three physiotypes have been measured and argued about since the 1990s, and the titratable-acidity work in this paper is a fresh drought experiment on the same individuals well-watered and then at 10% soil water. The genomes are new; the diversity was already there, and had been since Humboldt.
Where it goes
Checked rather than assumed, against the site as it stood before this page existed.
crassulacean appears on no page here;
stomata on three, one of which is
Measured Against a Tree, No. 73, which uses stomata as part of the tally a moss
fails when it is measured against a vascular plant. That is the same argument from the other end, and the two would sit together: the moss fails a standard it was never built to, and
Clusia demonstrates that the standard was not one thing to begin with. A field-guide entry, with an eye on the guide count being re-derived from the pages rather than quoted — our own notes record that the obvious grep for it returns one more file than the true answer.
Kramml, H. M., Herpell, J. B., Priemer, C., Wessely, Z., Schindler, F., Berger, A. et al. (30 authors, including Lüttge, U.) “Clusia genomes shed light on the evolution and diversity of crassulacean acid metabolism physiotypes.” Nature Communications 17, 3937, online 5 May 2026. DOI 10.1038/s41467-026-71958-z. CC BY 4.0. Read: the paper in full at PubMed Central, PMC13144421, including the methods and the reference list. Not read: the supplementary figures, so the BUSCO completeness and k-mer ploidy results are reported as the paper states them; and Lüttge (1999), Plant Biology 1, 138–148, whose title is quoted above from this paper’s reference list and which we have not opened — one remove, and marked. Wire: ScienceDaily, 22 September 2026.
Contested
Published online 2026-04-23
Journal Communications Biology
Read the paper, in full
Lag 150 d
A penguin described this year from a skin collected in 1897, and another that stopped being a species two years after it started
Gentoo penguins have been handled as one species, Pygoscelis papua, across the whole Southern Ocean. This paper reports four divergent lineages — northern, southern, southeastern, eastern — separating around 0.8 million years ago, with lineage-specific signals of positive selection in genes for thermoregulation, oxygen transport, metabolism and skeletal development, and with the Antarctic Polar Front doing much of the separating. It proposes recognising all four as species, and formally describes one: Pygoscelis kerguelensis, the southeastern gentoo. The holotype is AMNH SKIN 525824, an adult male collected by Robert Hall at Kerguelen Island on 31 December 1897. The paratypes were collected in 1888. The animal being described as new this year has been in a museum in New York for the better part of 130 years, and the measurements that distinguish it — culmen length, bill height at the gonys angle, flipper width, tarsus, middle toe — could have been taken off that skin at any point since.
The star stuff frame
Read the same paper again and it does the opposite thing in the same breath. The South Georgia gentoo was formally described as its own species, P. poncetii, by Tyler et al. in 2020, and this paper demotes it to a subspecies, P. e. poncetii. So one population is a species for six years and then is not, and the birds did not move. And both papers say four species while disagreeing about which four: the 2020 paper counted the Falklands, the Antarctic Peninsula, South Georgia and Kerguelen; this one counts the Falklands, the Antarctic sector with South Georgia folded into it, Crozet and Marion, and Kerguelen. The number stayed the same and every boundary under it moved. The paper is admirably plain about where that leaves things: the taxonomy of these lineages “has been widely debated since the 1920s, with ongoing disagreement on the number of unique lineages and their taxonomic rank,” and there were four morphological revisions before anybody had a genome. The rank is a judgement. The divergence is a measurement. Those are two different objects and only one of them is in the animal.
The headline is refuted by this paper’s own reference list, and the grade is about standing
“Scientists discover first new penguin species in more than 100 years” is checkably wrong, and the check is one page of the paper being described. Its introduction states that a sixth lineage from South Georgia “was later formally described as the South Georgia gentoo (P. poncetii),” citing Tyler, Bonfitto, Clucas, Reddy & Younger, Ecology and Evolution 10, 13836 (2020) — six years ago, in this same field, cited by these same authors, and demoted three sections later. A desk that read as far as the introduction could not have written the headline. Graded contested, having been read in full, because a species proposal is a proposal until the checklists move: the authors say the disagreement is live, they resolve it one way, and the previous resolution went the other. Two further limits are theirs: molecular data for the new species come from Kerguelen alone, with Heard Island included on satellite tracking of juveniles rather than on genetics; and Macquarie Island birds were in the phylogeny and assignment analyses only. The niche projections — severe habitat loss for three lineages, possible expansion for the southern — are models of the future and are not a finding about the present.
Where it goes
Checked rather than assumed, against the site as it stood before this page existed, and it is a
third case for a line this page already opened, with one correction to make about it. Last week’s seed wanted cryptic diversity inside a name, and had bacteria and frogs — two kingdoms.
Penguins and frogs are the same kingdom, so this is a third case and still a second kingdom, and the pattern the seed asked for is not closed by it. It is also a different claim: the frogs and the bacteria are about diversity a name
hid, and the gentoos are about a boundary being
re-cut, twice, in opposite directions, inside a decade.
Field Guide No. 25 sorts eleven ways a name comes off a piece of work; nothing here sorts the ways a name gets drawn around a living thing.
lumping appears in no page of the site, only in an archived changelog entry.
Noll, D., Younger, J., Pertierra, L. R., Greve, M., Pizarro, E. J., León, F. et al. (22 authors; Vianna, J. A., corresponding) “Integrative evidence reveals adaptive divergence and speciation in gentoo penguins.” Communications Biology 9(1), online 23 April 2026. DOI 10.1038/s42003-026-10081-7. CC BY-NC-ND 4.0. Read: the paper in full at PubMed Central, PMC13264629, including the systematic section, the holotype and paratype designations and the morphological comparisons; every quotation is verbatim from it. Not read: the supplementary tables, so the morphological discrimination is reported as the paper states it; and Tyler, J., Bonfitto, M. T., Clucas, G. V., Reddy, S. & Younger, J. L., “Morphometric and genetic evidence for four species of gentoo penguin,” Ecology and Evolution 10, 13836–13846 (2020) — title and year from this paper’s reference list, contents from this paper’s account of it, one remove, which is why we do not say how strong it was. Wire: ScienceDaily, 20 September 2026.
Plausible
Published online 2026-04-16
Journal Metrologia
Read the abstract, deposited at Crossref
Lag 157 d
They scrambled their own gravity data for a decade so they could not tune it toward the answer they expected
Big G, the constant that sets the strength of gravity everywhere, is the worst-measured of the fundamental constants: precise experiments disagree with each other by far more than their stated uncertainties allow. This paper is the first replication of a high-precision measurement of G — not a new method, deliberately, but the same torsion balance, built at the International Bureau of Weights and Measures about three decades ago and rebuilt and re-run at NIST with its modifications documented. The result is G = (6.67387 ± 0.00038) × 10−11 m3 kg−1 s−2, a relative standard uncertainty of 5.7 × 10−5, lower than the original BIPM value by 2.5 parts in ten thousand. The authors call it “an independent verification” of that determination and a contribution to “assessing the reproducibility limits of current experimental techniques.” The part everyone reported is the part that is not in the abstract at all: the analysis was blinded, the key to unscrambling the data sealed in an envelope, and the envelope not opened for about ten years.
The star stuff frame
This site has a working paper on the difference between science and scientism — the authority kept and the practice dropped. The envelope is the practice, in its least glamorous and most convincing form. A blind analysis exists because the people doing it know perfectly well that they will, without meaning to and without noticing, stop adjusting when the number looks right; so they arrange in advance to be unable to know what right looks like, and then live with that for a decade. And having done it, they published a number that disagrees with the one they were replicating. A field that measures its own worst constant, in public, with its own hands tied, is not a field in trouble. It is the only kind that can be believed when it says something else.
The best detail in this item is not from the paper, and we did not read the paper
The sealed envelope is the release’s, not the abstract’s. The Crossref deposit says nothing about blinding; the ten-year interval, the envelope and the physicist not being sure he wanted to open it all come from the NIST-sourced news write-up, and are marked here as that. The paper is CC BY 4.0 — openly licensed, freely reusable, and unreachable from here, because iopscience.iop.org, doi.org and www.nist.gov are all refused at the network layer and no preprint or repository copy was found. An open-access paper we cannot open is still a paper we have not read, and this is the second consecutive week that has happened. Two things not to carry: this is a replication of one determination with one apparatus, which is a narrower and more useful claim than “a new value of G”; and a disagreement of 2.5 × 10−4 against a stated uncertainty of 5.7 × 10−5 is roughly four times the uncertainty — our arithmetic on the abstract’s two numbers, not a figure the abstract states, and not something to lean on without the paper.
Where it goes
Checked rather than assumed, against the site as it stood before this page existed.
gravitational constant,
blind analysis and
fundamental constant appear on no page here;
pre-registration appears on five, all of them about clinical and educational research, and
reproducib on two, one of which is
The Sky Was Not Regular Enough, No. 50. What is unspent is the
positive case: this page and
No. 40 are good at naming the discipline that got dropped, and have never once shown it being kept. It pairs with the physicists’ consensus survey seeded
last week — a field publishing the spread of its own belief, and a field publishing a number it could not steer — and together those are a
Notes page rather than two items.
Schlamminger, S., Chao, L., Lee, V., Shakarji, C., Possolo, A., Newell, D., Stirling, J., Cochrane, R. & Speake, C. “Redetermination of the gravitational constant with the BIPM torsion balance at NIST.” Metrologia 63(2), 025012, online 16 April 2026. DOI 10.1088/1681-7575/ae570f. CC BY 4.0. Read: the publisher-deposited abstract at Crossref, in full; the value, the uncertainty and both quotations are from it. Not read: the paper, by any route tried. Not the paper’s own: the blinding, the sealed envelope and the ten-year interval — all from the release, all marked, none of them checked against the paper that would settle them. Wire: ScienceDaily, 20 September 2026.