A recurring, evidence-scored roundup of the newest science on the Frontier index.
Papers that entered the index after this issue are ranked on New this week, which updates every day.
35 new breakthroughs went live on Frontier between 2 and 8 November 2026, and biomedicine led them with 11.
Every entry in this issue is a primary research result traced to its source and scored from 0 to 100 on three axes: Evidence, Impact and Novelty. All 35 are peer-reviewed journal papers, 8 of them in Nature, 8 in PNAS and 5 in Science; 32 were published on or after 21 October.
Two results share the top score, 78. One points to a bacterial toxin as the driver of a subset of necrotising enterocolitis, the deadliest gut emergency in premature babies, and names the receptor it works through. The other sequences 712 ancient genomes and finds a population movement that, if it carried the language, would put Germanic's arrival in Scandinavia some 800 years later than tradition assumes. Two more follow at 77: an X-ray method that reads collagen strain at two scales inside an intact rat spinal disc, and a statistical model that finds nearly half again as many genes with detectable genetic effects in the same immune-cell data.
Section 2 takes three papers mapping genetic regulation one cell at a time; section 3 goes to the base of the tree of life and the chemistry between microbes and their hosts; section 4 collects four studies of the brain; section 5 gathers two searches that found nothing and one measurement that ruled an explanation out; section 6 takes the medicine. Section 7 explains why a Nature paper scored 78 while a Science paper in the same week scored 59.
Biomedicine and health led with 11 entries, then physics (6), neuroscience (5), AI and computing and space and astronomy (4 each), climate and environment (2), and one each in materials and energy, mathematics and quantum. Scores as read early on 9 November, before that morning's refresh: the six entries listed on 8 November have not yet had their first re-score, which can move a score by 2 or 3 points, and the index re-scores every entry as evidence moves, so a live page can differ.
1. The top of the table, in depth
Three of the four results at the top, in score order; the fourth, tied at 77, opens section 2.
Clostridium perfringens Delta toxin targets CD9 to drive NEC and brain injury
Frontier Score 78 · Biomedicine & Health · Nature
Necrotising enterocolitis is the most devastating gastrointestinal emergency in preterm infants, frequently fatal, and the gut microbiota had been implicated without a specific culprit. The authors find Clostridium perfringens carrying Delta toxin in the gut of infants with severe and fatal disease: anaerobic culture found the bacterium more often in patients than in infants without the disease (56% against 18%), and antibodies to the toxin turned up in 7 of 12 infants with severe surgical disease and in none of the infants without it. In neonatal mice, a toxin-positive clinical isolate, or purified toxin alone, reproduced both the bowel necrosis and the acute brain injury. A genome-wide CRISPR screen named the receptor: knocking the tetraspanin CD9 out of human cells cut the toxin's potency about 40,000-fold, deleting it in mouse myeloid cells protected the animals, and a neutralising antibody protected against disease progression. The human case is association in a small cohort, the causal chain is mouse work, and the authors note they had no toxin-deletion strain to confirm the pathology is toxin-dependent. Its Novelty, 91.7, is the week's highest. Frontier entry · Paper
Tracing the spread of Germanic languages using ancient genomics
Frontier Score 78 · Biomedicine & Health · Nature
When Germanic reached Scandinavia is contested: the cultures usually credited with bringing it predate the first written Germanic by 2 to 3 millennia. The authors sequenced 712 ancient genomes and analysed them with 4,009 published ones, and find a population hard to see in the archaeological record that entered Sweden, likely from the Baltic, by around 4000 years before present and was widespread in Scandinavia by 3500. By 1650 years before present that ancestry moves south into presumed Celtic-speaking areas, and during the Migration Period, 1575 to 1375 years before present, it appears among Germanic-speaking Anglo-Saxons in Britain and Langobards in southern Europe, before a related back-migration north into Denmark and southern Sweden in step with the documented emergence of Old Norse. That would place the arrival some 800 years later than traditionally assumed. This is ancestry, not language: the link rests on matching movements to the stages linguists date, the authors call the Baltic-linked group a potential vector and cannot rule out cultural diffusion in the opposite direction, and the group's proposed origin, somewhere between Finland and northeast Sweden, is a region from which no genomes yet exist. Frontier entry · Paper
Multimodal X-ray imaging reveals hierarchical fibre mechanics
Frontier Score 77 · Physics · Nature
The outer ring of a spinal disc is built from plywood-like layers of collagen fibres, themselves built from nanometre-scale fibrils, and how the two scales share load inside an intact disc was unresolved. The authors introduce TomoSAXS, a full-field three-dimensional X-ray scattering tomography that, with phase-contrast tomography, maps fibril-to-fibre mechanics across whole tissue. On 3 intact rat discs imaged before and after a 50 μm compression it reconstructed nanoscale structure for more than 55% of all fibres in the scanned volume, and in one representative disc found fibril strain falling as the molecular pre-strain inside fibrils rose, which the authors read as load-sharing through molecular unwinding. Fibre strain rose with fibre curvature in that same disc, which they suggest may partly explain why the posterior regions of the disc are most prone to herniation. It is a methods paper on few samples: three rat discs under a single compressive step and one fixed human disc, with modest individual regressions, and the herniation link is a suggestion rather than a clinical finding. Frontier entry · Paper
2. Reading genetic regulation one cell at a time
Three papers this week, scoring 77, 75 and 69, ask what a gene or a disease-linked variant actually does, and where. Two of them build their answer from single cells rather than ground-up tissue; the third knocks genes down one at a time in human neural cells.
Efficient and accurate mixed model association tool for single-cell eQTL analysis
Frontier Score 77 · Biomedicine & Health · Nature Genetics
Mapping how variants change gene expression in single-cell data usually means averaging each donor's cells into one pseudobulk profile, which throws information away. SAIGE-QTL is a Poisson mixed model that works on the single-cell read counts directly. In 1.2 million immune cells from 982 people it identified 48.8% more genes with a significant expression QTL than the pseudobulk pipeline it was compared with (17,218 against 11,569, counted per cell type), found 5,541 rare-variant eGenes across 14 cell types, and ran 244-fold faster than the model it matches. The honest caveats are in the paper: the pseudobulk tools still ran about four times faster than SAIGE-QTL, a Poisson model may fit overdispersed genes less well than a negative binomial one, and more eGenes means more detected associations, not identified causal variants. Frontier entry · Paper
Population-scale single-nucleus multi-omic profiling of skeletal muscle
Frontier Score 75 · Biomedicine & Health · Nature Genetics
The same question, asked of muscle. Pooling and genetically demultiplexing 287 frozen human muscle biopsies gave single-nucleus maps of both open chromatin and gene expression across 13 cell types, yielding more than 7,000 expression QTLs and more than 100,000 chromatin accessibility QTLs. Almost 3,500 genome-wide association signals across 38 traits colocalize with them, most in a cell-specific way, and in the preprint 80% of those were chromatin rather than expression colocalizations, partly because far more chromatin QTLs were detected. The causal direction maps are statistical inference, not experiment, and the authors write that confirming the impact on disease remains a critical step for future studies. Frontier entry · Paper
Transcriptomic convergence and the female protective effect in autism
Frontier Score 69 · Neuroscience · Nature
Hundreds of genes are implicated in autism, each individually rare, which leaves the question of how such different genes produce related traits. Knocking down 18 autism genes one at a time in human neural progenitor cells changed a median of 920 genes each, and the changes converged: 27 high-confidence autism genes were altered by at least 10 of the knockdowns, all of them downregulated. The reconstructed regulatory network puts CHD8 and candidates including REST at the centre, and the X-linked factor ZFX, expressed from both X chromosomes in females, emerged as an activator of many autism genes, which the authors propose as a buffer behind the female protective effect. That proposal was not tested directly, the work is in one male stem cell line read over a short window, and the regulators were identified computationally. Frontier entry · Paper
3. Deep relatives, and chemistry between kingdoms
Eukaryotes' closest relatives are internally simple syntrophic peptidotrophic archaea
Frontier Score 74 · Biomedicine & Health · Nature
Eukaryotic cells are thought to descend from archaea of the phylum Promethearchaeota, but the cultured representatives were distant relatives, so nobody had seen the cell biology of the lineage nearest to us. The authors isolated two strains from Heimdallarchaeia, the lineage branching closest to eukaryotes, and grew them with methanogenic partners. Both are obligately anaerobic peptide degraders that make protrusions and vesicles but have no persistent internal membranes, and both grow extraordinarily slowly: one doubles in about 7 to 12 days, the other in about 14 to 20. One carries a cytochrome c oxidase, a gene associated with aerobic life, yet its cultures cannot grow in oxygen unless aerobes are present, which the authors read as oxygen detoxification rather than respiration. That argues against the idea that a complex internal membrane system came first, before the mitochondrial ancestor. The inference about an ancestor billions of years old rests on how widely these features are conserved across the phylum, not on a fossil. Frontier entry · Paper
Steroid hormones produced by coral-associated Endozoicomonas influence coral immune response during thermal stress
Frontier Score 71 · Biomedicine & Health · Nature Microbiology
Endozoicomonas bacteria live inside corals everywhere and track with coral health, but their contribution had been read off gene lists. Sampling a coral in the Arabian Gulf across 32 °C, 34 °C and 27 °C, the authors found the family these bacteria belong to dominating healthy tissue at 60% of the bacteria on average and falling to 1.3% at disease lesions, with the Endozoicomonas genome itself making up 35 to 43% of the genomes recovered from healthy tissue and 0.8% at lesions. In culture, three isolates from different hosts and oceans converted cholesterol into terminal steroid hormones, testosterone or progesterone depending on strain and temperature. Giving coral fragments 10 nM of either hormone through a heat ramp amplified the heat-induced immune response across 39 innate immunity genes, raising survival signalling while suppressing an adaptor tied to inflammation; the same steroids at ambient temperature changed nothing. The chain is assembled from pieces: the field data are correlative, the conversion was shown in culture rather than inside a coral, and the experiments measured transcription, not survival. Frontier entry · Paper
Planetary-scale metagenomic analysis reveals CRISPR targeting dynamics across ecosystems
Frontier Score 66 · Biomedicine & Health · Nature Microbiology
A CRISPR spacer is a piece of DNA a microbe kept from an invader, so a catalogue of spacers is a record of what the planet's microbes have met. By pulling spacers straight out of short reads rather than assembled metagenomes, the authors collected 794 million spacers from 367,609 public metagenomes and matched them to 2.5 million viruses and plasmids, 1.18 billion hits in all. Spacers matching known viruses recur across samples while the rest turn over quickly, consistent with selection for keeping a record of phages, and matches to phages of other hosts are, in the authors' wording, evidence suggestive of phage DNA repeatedly landing in cells that are not its hosts. Frontier entry · Paper
4. Four ways into the brain
Connectome simulations identify a central pattern generator for fly walking
Frontier Score 71 · Neuroscience · Nature
Central pattern generators produce rhythms without receiving one, and no walking pattern generator's circuit structure was known in any animal. The authors turned the fly's leg-controlling connectome into a dynamic simulation and screened the descending command neurons; pruning the network isolated a minimal rhythm-generating circuit of three interneurons, one inhibitory and two excitatory, reported as necessary and sufficient for rhythms across all six legs in four connectome datasets. Two of the simulation's predictions held up in living flies with optogenetics. The circuit itself is a model result, which is why the authors call it putative: nobody has silenced any of the three cells in a fly, and in the preprint several muscles that should be active in walking came out silent, which the authors read as the circuit being perhaps enough to initiate stepping, while the naturalistic pattern likely relies on more. Frontier entry · Paper
Aggression experience and observation drive shared behavioural and neural changes
Frontier Score 70 · Neuroscience · Nature
Male mice trained for 8 days by fighting, by watching fights from a platform, or by spending matched time with a toy were recorded from 12 sites across the brain's social behaviour network. In a later harder fight, the mice that had only watched and the mice that had fought shared a behavioural strategy the controls lacked, their neural dynamics grew more similar, and both showed reduced inhibitory activity across regions. In a separate cohort of observers, trained by watching for 5 days, chemogenetically restoring inhibition in four of those regions 30 minutes before the fight made them behave more like untrained mice, which the authors take as showing the reduction is necessary for the change. The decoupling the abstract describes is a statistical result rather than a measured synapse, the groups were small and male-only, experimenters were not blinded during the hard fight, and the step to human aggression is the authors' suggestion. Frontier entry · Paper
Transcranial magnetic stimulation to the dorsolateral prefrontal cortex modulates single-neuron activity in humans
Frontier Score 68 · Neuroscience · PNAS
Prefrontal magnetic stimulation is an established depression treatment, and what it does to single neurons in the deep circuits that regulate mood was unanswered. Recording 185 neurons in four neurosurgical patients, the authors found a single pulse produced time-locked responses in 46% of them, in deep cortical and subcortical structures on both sides of the brain. Putative inhibitory neurons in striato-thalamic circuits tended to fire more from about 8 ms, while putative excitatory neurons and limbic areas were suppressed on a slower course. The evidence comes from four people with medication-resistant epilepsy, neuron sampling was sparse and uneven, the network correlation comes from one participant, and the stimulation was single pulses rather than the repetitive protocols used in treatment. Frontier entry · Paper
Identification of candidate markers for the isolation of neuron-specific extracellular vesicles
Frontier Score 69 · Neuroscience · Nature Neuroscience
Every cell sheds vesicles carrying its own proteins, so a blood test that captured neuronal ones would read tissue no biopsy reaches. The field's most-used neuron marker, L1CAM, was shown by these authors to circulate as a free protein instead. They set three requirements for a candidate and built a purification that improved vesicle purity more than 100-fold over size-exclusion chromatography, catalogued 2,104 vesicle proteins in cerebrospinal fluid and 1,862 in plasma, and crossed them with neuron-specific expression. Of the three candidates they could assay, NRXN3 was confirmed on vesicles in both fluids, found in all 36 samples imaged, on about 4% of plasma vesicles. It is a proof of concept for isolation, not a biomarker result: no disease signal is reported, and NRXN3's soluble isoforms mean the antibody may capture protein that is not on a vesicle, the exact trap L1CAM fell into. Frontier entry · Paper
5. Two searches that found nothing, and one explanation ruled out
A null result is a real result, and this week produced two of them at different scales, alongside a measurement that closes off one explanation of a decades-old puzzle. Each narrows where an answer can be.
Any-light-particle searches with ALPSII: First science results
Frontier Score 69 · Physics · Physical Review Letters
ALPS II at DESY shines laser light through a magnetic field at a wall: if axions exist, light would partly convert into them, they would pass the wall, and some would convert back into photons behind it. In its first campaign, February to May 2024, it saw nothing, and set a limit on the axion-photon coupling of 1.5 × 10⁻⁹ GeV⁻¹ at 95% confidence for masses below about 0.1 meV, more than a factor of 20 better than all previous similar experiments. The value of a laboratory bound is that it does not depend on astrophysical or cosmological assumptions, unlike limits drawn from the Sun. The campaign ran with only the regeneration cavity in place, and the preprint reports a broad peak likely caused by stray light passing around the wall; the optics are now being upgraded toward another two orders of magnitude. Frontier entry · Paper
Modified Newtonian Dynamics and galaxy clusters
Frontier Score 62 · Space & Astronomy · PNAS
Galaxy clusters are the standing challenge to MOND, the proposal that modifies gravity instead of adding dark matter. Across 106 clusters this analysis maps how much hidden mass MOND still needs: 4 to 10 times less than Newtonian gravity requires, but not zero. The residual is concentrated in cluster cores and scales with the clusters' gas at roughly half its mass, which is the profile any ordinary explanation, stellar remnants or cold molecular clouds, would have to match. Nothing here detects such matter: the masses rest on hydrostatic equilibrium and spherical symmetry, and the paper reports where the missing mass would have to be, not that it is there. Frontier entry · Paper
Equivalence of pseudogap and pairing energy in a cuprate high-temperature superconductor
Frontier Score 59 · Physics · Science
The pseudogap is the energy gap that opens in cuprates above the superconducting transition, and what it is has been argued for decades. Shot-noise measurements here read the charge the current carries and find electron pairing setting in at the pseudogap energy, with a slope of 1 between the two across positions in the preprint. That excludes a purely local-order origin and ties the pseudogap to pair formation. Because pairs form at such large energies, the authors suggest what limits the critical temperature of cuprates is phase coherence rather than pairing strength. Pairing is inferred from the charge rather than seen, and the measurements are at 4.2 kelvin and below on underdoped and optimally doped samples. Frontier entry · Paper
6. Medicine: a model, a valve, an outbreak and a forecast
An open vision-language model for diverse medical applications
Frontier Score 74 · AI & Computing · Nature Medicine
A hospital cannot send patient records to a model it does not control, which is the practical case for open weights. MedGemma is a family of medical vision-language models built on Gemma 3 at 4 and 27 billion parameters, with a 400-million-parameter medical image encoder. On held-out tasks it improves on its base by 2.6 to 10% in medical image question answering and 15.5 to 18.1% in chest X-ray finding classification, and scores 86.2 on MedQA. A single thoracic radiologist, who knew which reports were AI-generated, rated 68% of its reports on normal chest X-rays and 49% of those on abnormal ones as equal to or better than the original. These are benchmarks rather than clinical validation: some are near saturation, older public ones may have leaked into training data, the report ratings come from one unblinded reader, and there is no support for CT or MRI series. Its Impact pillar, 93.8, is the week's second highest, behind the cosmic-dawn black hole at 95. Frontier entry · Paper
TAVR in Patients with Moderate Aortic Stenosis with At-Risk Features
Frontier Score 61 · Biomedicine & Health · New England Journal of Medicine
Guidelines currently recommend watching a moderately narrowed aortic valve rather than replacing it. In 744 patients with at-risk features, replacement cut the composite of death or heart-failure events at 2 years to 16.7% from 23.9%. The gain came from heart-failure events, 19.0% down to 10.8%; deaths were not lower, 8.4% against 7.3%. The trial was open-label and funded by the valve's manufacturer, its patients averaged 78 years, and two years says nothing yet about the durability of valves implanted earlier than usual. Frontier entry · Paper
Genomic analysis of early case samples from the 2025 Marburg virus outbreak in Ethiopia
Frontier Score 64 · Biomedicine & Health · Nature Microbiology
Ethiopia's first Marburg outbreak, November 2025, killed 64% of confirmed cases. Seven near-complete genomes from seven early cases differ from each other by only 0 to 6 single-nucleotide polymorphisms, which the authors read as human-to-human transmission following a single spillover, a finding that points containment at contacts rather than at the bat reservoir. The sequencing was done in Addis Ababa, and the authors say it informed the deployment of remdesivir and a vaccine during the response. The dataset is thin: seven of the 14 confirmed cases, with broad and overlapping credible intervals on the dating. Frontier entry · Paper
Global, regional and national burden of cervical cancer in 1990-2023 with forecasts to 2050
Frontier Score 61 · Biomedicine & Health · Nature Medicine
Cervical cancer is highly preventable, and the forecast is that it will kill more women, not fewer: deaths are projected to rise 39.3% from 2024 to 2050. It killed an estimated 369,000 in 2023, with age-standardised incidence and death rates more than four and seven times higher in low-income countries than in high-income ones. These are model estimates with wide uncertainty: the 887,000 new cases in 2023 sit within an interval of 682,000 to 1,170,000, and several trends cannot be told apart from no change, including the forecast changes in the rates. Frontier entry · Paper
7. Where the points went
One of the week's two top entries, the Delta toxin paper, scored 78. The cuprate pairing paper scored 59. Both are in flagship journals, both are days old, and both keep more than 99% of their pillar-weighted standing after the recency factor, so the 19-point gap is in the pillars.
- Evidence, 91.9 against 91.0. Both peer-reviewed in a flagship venue, both open access, neither carrying a retraction or concern.
- Impact, 39.3 against 49.6. The cuprate paper leads, because citations have begun to reach its record and none has yet reached the toxin paper's, which was published on 28 October, a week before it. Both entries show their live counts on their own pages; Impact reads what a record carries today, so for work this new it is mostly a placeholder.
- Novelty, 91.7 against 48.8, and this is the whole gap. Weighted at half the score, it is worth 45.9 points to the toxin paper and 24.4 to the cuprate one.
The sub-metric behind that is conceptual novelty, and it is worth saying what it measures, because the obvious reading is wrong. It is not how many papers made the same pairing of research topics before. Each entry's score breakdown names its median topic pairing and the number of earlier works that made it, and on those two pages the counts are close: 15,065 earlier works for the toxin paper's pairing, 19,134 for the cuprate paper's. Yet the same breakdowns read 100 and 4.3 on this sub-metric.
The measure is pointwise mutual information: how often two topics appear together against how often each appears at all. Both of the toxin paper's topics are vast literatures, so 15,065 joint papers is far fewer than chance would give. The cuprate paper's two topics are smaller and already deeply intertwined, so 19,134 joint papers is far more than chance would predict. An absolute count says nothing on its own; the ratio to the expected count is the signal. The Germanic genomics paper sits in between, on a pairing 555 earlier works had made, and its breakdown adds the second, smaller signal: the depth of the literature it builds on, a median reference age of 7.5 years, which lifts it to 72.5.
None of this says a decades-old question in superconductivity matters less than a toxin, or that a global cancer forecast matters less than either. The score asks a narrower question, how new and how well established a result is and how strongly it is being cited so far, and it deliberately rewards results that join fields rarely joined. A paper that answers a long-standing question inside one field is exactly the kind of work it scores conservatively, which is why the pillars are published next to every entry.
8. Also new this week
The other 15 entries, by field, with each one's score and journal; each title links to its Frontier entry, and each ends with a link to the paper.
Biomedicine & Health
- Rapid longitudinal flights cause time shifts and Sun compass error (70, Science). GPS tracks of homing Manx shearwaters show that flying far enough east or west shifts a bird's internal clock relative to local time, and their courses are then deflected in the direction a mismatched Sun compass predicts; they keep using it along coastlines where other cues are available. It is observational, so the finding is that free-flying birds' deflections match the prediction, weighed against rival models. Paper
Physics
- Defining and measuring proximity to criticality based on univariate time series (68, PNAS). A temporal renormalization group theory gives an information-theoretic way to estimate how close a system sits to the edge of instability from a single recorded time series, without knowing or touching its control parameters; applied to cortical spiking in 8 rats and 13 mouse recordings, it puts cortex closest to criticality awake and furthest in deep sleep. What is estimated is a property of the fitted autoregressive model, and the authors name bursty, strongly non-Gaussian dynamics as the failure mode. Paper
- Hydration at the base of the mantle transition zone by ancient subductions in Asia (65, PNAS). Combining mineral physics with seismic velocity and the depth of the 660-km discontinuity gives an average water content of about 0.14 wt% at the base of Earth's largest potential water reservoir, rising above 0.5 wt% beneath Asian belts where ancient oceans were subducted, which ties water sitting at 660 km today to subduction reconstructed back to 410 Ma. The water content is inferred from published seismic models rather than sampled. Paper
- Seafloor geodesy delineates fast megathrust relocking after the giant 2011 Tohoku earthquake (62, Science). The fault that produced the 2011 Tohoku earthquake appears to have regained a pre-earthquake-like locking state within 14 years. Locking is not observed directly: it is inferred by combining seafloor and onshore geodesy, repeating earthquakes and numerical modelling, and the paper was not openly readable, so this entry rests on the abstract alone. Paper
AI & Computing
- Intrinsically motivated humans and agents in open-world exploration (72, PNAS). Adults, children aged 8 to 12 and AI agents explored the same open-ended game, and three measures AI researchers use to make agents curious all tracked human progress; across both, diversity of states visited rose early and then plateaued while influence over the environment kept rising, which the authors read as novelty first, then control. The sample is small, the measures are correlational, and the private-speech result is explicitly exploratory. Paper
- Denoising the US Census: Succinct block hierarchical regression (65, Science). The 2020 US Census adds calibrated noise to protect confidentiality, then reconciles the noisy measurements into published tables; BlueDown redoes that reconciliation and cuts the error of county-level groupings by race by 21 to 36% and of tract-level groupings by housing type by 38 to 61%, at no cost to confidentiality, since it starts from the same released measurements. True counts are confidential, so accuracy is measured against the Bureau's published microdata as a stand-in. Paper
- Emergence of polarization in networks of large language model agents (65, Nature Communications). Networks of 1,000 LLM agents talking about political issues moved away from their near-Gaussian starting distribution in fewer than 10 rounds and polarized, the neutral share falling from 40% to between 0.4% and 22.5% depending on the issue, while the agents clustered with the like-minded. Every intervention tested reduced polarization, with individual-level ones working best; it is a simulation of language models, which the authors say must be validated against human behaviour. Paper
Space & Astronomy
- A Martian meteorite that formed ≳4.1 billion years ago samples Mars' early crust and mantle (67, Science). Most Martian meteorites are geologically young while the planet's surface is ancient, so a rock that formed at least 4.1 billion years ago samples a part of Mars the collection had missed, from a mantle source with a primordial composition; its altered minerals and oxygen isotopes record reaction with fluids before an impact ejected it. One rock shows that such crust existed, not how much of Mars resembles it, and the age is a minimum. Paper
- A gas-enshrouded and gas-reddened black hole at cosmic dawn (67, Nature). A source seen 660 million years after the Big Bang shows one of the largest hydrogen Balmer breaks reported at any redshift, at 7.7, beyond what stars are expected to produce, which the authors model as a black hole wrapped in dense, dust-free gas. If the redness comes from gas rather than dust, masses of such sources may be overestimated by orders of magnitude. The authors call their modelling highly simplistic. Paper
- A double debris disc around a white dwarf sculpted by two planetary bodies (63, Nature Astronomy). All 21 previously known white dwarf gas discs are single rings; this one is two, which the authors attribute to two surviving planetary bodies, each feeding a ring, and the geometry bounds how large such bodies can be. The two bodies are proposed rather than detected, and the authors weigh alternatives. Paper
Climate & Environment
- Thermal ecology drives alternate responses to rising temperatures in Caribbean anoles (66, PNAS). A decadal resurvey of anole lizards on three Caribbean islands finds that species in thermally varied open habitats buffer heat behaviourally and gained little heat tolerance, while species of thermally uniform closed forest, poor at thermoregulating, raised their heat tolerance by 2 °C on average; thermal ecology, not relatedness or proximity, is the key factor. Acclimation and rearing experiments point to both rapid evolution and plasticity without apportioning between them. Paper
- Global bee bioregions mirror those of other taxa despite a uniquely bimodal latitudinal diversity gradient (63, PNAS). Bees are richest at mid-latitudes rather than in the tropics, yet the boundaries between their faunas, 21 bioregions in at least six realms drawn from 3,370 species, fall close to those of plants, tetrapods, ants and butterflies, delimited by the same ancient barriers. That separates where biotic boundaries sit from how many species live inside them, which the authors read as distinct processes. Paper
Materials & Energy
- Atomically resolved exciton emission from single defects in MoS2 (67, Nature Nanotechnology). Light excited by a scanning tunnelling microscope's own current, with the monolayer decoupled from its substrate, maps emission defect by defect and ties a defect glow common to MoS2 to negatively charged carbon-hydrogen complexes rather than to the sulfur vacancies often credited; cleaving off the hydrogen with a voltage pulse left a carbon defect with its own spectrum. Measurements were below 5 kelvin in ultrahigh vacuum, and the step to engineered quantum emitters is an outlook. Paper
Mathematics
- The ABCD hypothesis (66, PNAS). Cantor's question about the size of the reals is one case of comparing sets of functions between infinite cardinals, and the axioms with choice leave it undetermined; the ABCD hypothesis gives every such comparison the simplest possible answer, and this paper sketches a proof that it holds below Theta when the Axiom of Determinacy holds and the universe is L(R). Determinacy contradicts the Axiom of Choice, so this says nothing about the universe most mathematicians work in, and the article describes itself as a sketch. Paper
Quantum
- Conserved quantities enable the quantum Mpemba effect in weakly open systems (63, Physical Review Letters). For a weakly open quantum system started in a thermal state, extra conserved quantities appear necessary, though not sufficient, for a hotter system to relax faster than a colder one; in simulations, chaotic chains with only energy conserved relaxed smoothly while integrable chains showed the effect. It is theory supported by numerics, argued at a zeroth-order approximation in the coupling, with the largest chains using that approximation itself. Paper
9. What the week says
A week computed rather than measured. Much of this week's work gets its result from data that already existed or from a model standing in for an experiment. The fly's walking circuit comes out of a connectome simulation, the proximity to criticality out of fitted autoregressive models, the mantle's water out of published seismic models, MOND's residual mass out of X-ray maps, the census gains out of reprocessing released measurements, and the polarization result out of language models talking to each other. Each paper is explicit about it, and in each case the test that would turn the inference into a measurement is clear enough to name: silencing the three fly interneurons in a behaving fly, applying the criticality measure to a system whose control parameters can actually be turned, finding the ordinary matter MOND's residual profile would require.
Nulls and bounds. ALPS II saw no axions and tightened the laboratory limit more than twentyfold, MOND still needs unseen mass across 106 clusters, and in weakly open quantum systems the Mpemba effect appears to need extra conserved quantities, which a chaotic chain does not have. None of the three is a discovery of something new, and all three narrow where the next experiment should look, which is the part of science that rarely makes a headline.
Where a gene's effect shows up. Three independent teams, in immune cells, in muscle and in autism gene networks, spent the week on the same problem: a gene or a trait-linked variant acts somewhere specific, and the usual way of measuring hides where. Two answered by reading single cells instead of ground-up tissue; the third by knocking 18 autism genes down one at a time and watching which other genes moved. All three stop at association or at cultured cells, and all three say so.
Mice, dishes and small cohorts. The toxin behind a subset of necrotising enterocolitis, the observational learning of aggression, and the convergence of autism genes are mouse and cell-culture results with at most a small human cohort beside them. They are where the next experiment starts, and the results most easily carried to patients too soon.
10. How these scores are built
The Frontier Score weighs Evidence (25%), Impact (25%) and Novelty (50%), then applies a recency factor under which a result keeps 75% of its standing at eight months old and never falls below half, so the active edge stays on top. A retraction collapses the score to zero, and a separate confidence label says how settled each result is: every entry in this issue reads Early signal, because their citation histories have barely begun.
The index takes in up to six new papers a day from leading journals, first the papers their own journal ran a commentary on, then the strongest of the rest. The full recipe, its sources and its limitations are on the methodology page, and our guide to the replication crisis explains how to judge a result this new. To see where any paper stands on citations, inside the index or not, our free paper check shows its field-weighted citation impact and its citations in five open indexes, the list check does the same for up to 25 papers at once, and the researcher check reads a whole publication record from an ORCID iD.