Skip to content

27 min read

Science Breakthroughs This Week: 9-15 November 2026, Scored

The science of 9 to 15 November 2026, scored: 40 new results, led at 76 by a supersymmetric reading of an unexplained dark-matter recoil, with three altermagnet papers in one week.

By The Frontier Desk. Scores quoted are Frontier Scores as of the publication date above, and each entry's page shows its live score; see how they are computed.

The Frontier Brief. The week's top new results, scored, by email every Monday. What you get

One email a week. The top new breakthroughs, scored. Unsubscribe anytime.

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.

40 new breakthroughs went live on Frontier between 9 and 15 November 2026, and biomedicine led them with 16.

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 40 are peer-reviewed journal papers: 10 in Nature, 5 in Physical Review Letters, 3 each in Science and the Lancet, and the rest across Cell, NEJM, Nature's specialist titles and the Physical Review journals.

The week's top entry, at 76, is not a measurement at all but a reading of one: a proposal that an unexplained recoil event in a leading dark-matter detector could be a Higgsino, the supersymmetric partner of the Higgs boson. Behind it, at 73, a large test of whether AI can generate research ideas scientists find interesting, with a sobering answer. Two results share 71: a map of how repeated stretches of DNA regulate immune genes cell type by cell type, and a cross-species atlas of the primate basal ganglia.

Three of this week's physics and materials papers are about altermagnets, a magnetic phase only recently named, which get their own section. Section 3 reads a gene's effect where it acts, section 4 takes the brain and a settled quantum limit, section 5 gathers cross-checks and bounds, and section 6 takes the medicine. Section 7 explains why the week's most-cited paper is not its top-scored one.

Scores are as read early on 17 November, before that morning's refresh; the six entries listed on 15 November have had only their first re-score, and the index re-scores every entry as evidence moves, so a live page can differ.

1. The top of the table

Higgsino dark matter interpretation of the LUX-ZEPLIN 248 keV nuclear-recoil event

Frontier Score 76 · Physics · Physical Review Letters

The LUX-ZEPLIN experiment, a xenon detector deep underground, recently reported a single nuclear-recoil event it could not attribute to any known background, at 248 keV, in tension with a background-only hypothesis at a global significance of 2.6 sigma (local 3.4 sigma). This paper asks whether a Higgsino could produce it. A nearly pure Higgsino is a form of inelastic dark matter, two nearly identical neutral states separated by a small mass splitting that scatter off a nucleus by converting one into the other, and because the same electroweak interaction sets both the coupling and the rate, the model has almost nothing to tune. The authors show that for a Higgsino mass near 1 TeV and a splitting near 350 keV, the predicted cross section approaches the interval LZ reported for the event. It is a consistency argument, not a detection: at 3.4 sigma the event is far below a discovery, and a single event cannot establish a particle. The closest match is at exactly 1 TeV; the mass thermal production prefers, near 1.1 TeV, falls just outside their rough rescaling of the LZ limits, which they note a slightly larger splitting would close, and they stress that a proper limit needs a dedicated LZ analysis of the full recoil spectrum rather than their rescaling. Frontier entry · Paper

Generating interesting scientific ideas using knowledge graphs and LLMs

Frontier Score 73 · AI & Computing · PNAS

Can AI generate research ideas that working scientists find interesting? The authors built SciMuse, which draws personalized ideas from a knowledge graph of 58 million papers and a large language model, then ran the evaluation that matters: more than 100 research group leaders rated over 4,400 ideas for how interesting they were. The answer was modest. The mean rating was 2.40 on a 5-point scale, the most common rating was 1, and 24.9% of ideas were rated 4 or 5. The knowledge-graph step meant to be the method's advantage did not beat the same GPT prompt given only the two researchers' recent paper titles, and pairs the graph predicted would be high-citation even tilted weakly toward lower interest. The authors frame the contribution as a methodology plus a large interdisciplinary expert evaluation, and show that interest can at least be predicted, by a supervised network and by a zero-shot LLM ranking, above a coin flip. Interest is a judgement, not an outcome: the study never followed up on whether any idea was pursued, the raters are a narrow slice of science, and whether selecting ideas with these predictors improves what is generated remains untested. Frontier entry · Paper

Tandem repeat variation shapes immune cell type-specific gene expression

Frontier Score 71 · Biomedicine & Health · Nature

Tandem repeats, short DNA sequences repeated back to back, are highly variable between people and hard to genotype, so studies that read single letters usually skip them. The authors paired whole-genome and single-cell RNA sequencing from more than 5.4 million blood cells across 1,925 people to ask how variation at more than 1.7 million repeat sites affects gene expression in specific immune cell types. They find more than 69,000 single-cell expression QTLs, 30.7% of them specific to one of 28 cell types, and matched chromatin data from more than 3.4 million nuclei show accessibility effects for nearly a third of the expression-associated repeats. Fine-mapping implicates 1,490 repeats as candidate causal drivers in 6.1% of genes tested. These are statistical associations in blood, not demonstrated mechanisms; repeat genotyping from short reads is imperfect (80.9% concordance against long reads in the preprint), and both cohorts are of European ancestry. Frontier entry · Paper

Cross-species multiomic atlas of the primate basal ganglia

Frontier Score 71 · Neuroscience · Cell

The basal ganglia sit at the centre of movement disorders and psychiatric illness, but a cell-type map shared across species has been missing. The authors assembled a consensus atlas of 1.8 million nuclei from humans, macaques and marmosets across eight basal ganglia structures, combining gene expression, open chromatin and spatial location, and defined 61 homologous cell types conserved over 76 million years of primate evolution. A cell's dorsal-ventral position was more conserved than its D1/D2 identity, and they describe a distinct D2 spiny projection neuron type with its own molecular and electrical signature. Mapping genetic risk onto the atlas concentrated psychiatric risk in ventral striatal spiny projection neurons. The atlas is built from a limited number of postmortem donors per species (4 to 9), so it is a reference and a hypothesis-generating resource rather than a powered test, and the genetic-risk mapping is statistical enrichment, not a demonstration that a cell type causes a disorder. Frontier entry · Paper

2. Three altermagnet papers in one week

Altermagnetism is a magnetic phase named only in the past few years: a material with no net magnetization, like an antiferromagnet, but whose electronic bands split by spin with momentum, like a ferromagnet. Three of this week's entries probe it, and together they show a young field moving from prediction to measurement.

Chiral altermagnon in MnTe

Frontier Score 68 · Physics · Science

Altermagnets were predicted to carry spin waves with a handedness fixed by their symmetry. The authors give direct experimental evidence for one in the altermagnetic candidate manganese telluride, using resonant inelastic x-ray scattering that compares how the material scatters left- and right-handed circularly polarized light. The handedness is read straight off the difference between the two channels; the magnon sits near 35 meV, and the dichroic signal changes sign across a nodal region, as the material's symmetry requires. What rests on symmetry is the altermagnetic reading of how the signal varies with momentum, and the two handed modes were never seen to split in energy: the measurement, one synchrotron beamline at 30 kelvin with a resolution near 21 meV, is far too coarse to separate two branches a few meV apart, so any splitting is argued from line widths. Manganese telluride is still called a candidate, so this is evidence for the phase rather than a settled classification. Frontier entry · Paper

Time-reversal-symmetry-breaking spin injection from a compensated magnet

Frontier Score 67 · Physics · Nature

Electrically injecting spin has conventionally needed either a ferromagnet's net magnetization or a relativistic effect. Here the authors inject spins from Mn5Si3, a magnet whose net magnetization vanishes, into a non-magnetic channel and detect their arrival downstream; switching between the material's time-reversed magnetic states reverses the signal, and the behaviour appears in two biasing geometries that differ from conventional injection. Symmetry analysis and first-principles calculations tie this to altermagnetic order. The quoted spin polarizations (6.5 to 10%) and diffusion lengths (50 to 130 nm) are calculated or fitted, not directly measured; what is measured is a non-local voltage that reverses with the magnetic state. The authors bound a spontaneous net moment at no more than 25 milli-Bohr-magnetons per manganese atom and call the work a proof of concept. Frontier entry · Paper

Altermagnetic signatures and topological textures in ultrathin multiferroic BiFeO3

Frontier Score 66 · Materials & Energy · Nature Materials

Multiferroics, with both electric and magnetic order, promise low-power switching, but both orders usually degrade in the ultrathin films devices need. The authors realize a multiferroic phase of BiFeO3 with no magnetically dead layer and report signatures of altermagnetism down to four unit cells, about 1.6 nm, where short-circuit boundary conditions and strain stabilize a collinear phase with d-wave altermagnetic order. As the title says, these are signatures, not a direct observation: the spin-resolved photoemission does not resolve separate spin-dependent bands, and the altermagnetic order is suggested by first-principles calculations with supporting evidence from a spin-polarization reversal. Their attempts to drive the band splitting by ferroelectric switching are not yet conclusive, because the heterostructure depoles rapidly. Frontier entry · Paper

3. Reading a gene's effect where it acts

A spatiotemporal cancer cell trajectory underlies glioblastoma heterogeneity

Frontier Score 68 · Biomedicine & Health · Nature

Cancer cells in glioblastoma take varied, changeable states, and whether those states follow a defined path was unclear. The authors built a single-cell and spatial atlas of 12 tumours (57 sites, more than a million nuclei) and find the variety within a tumour driven by a recurring, spatially patterned transition from development-like states toward states defined by a glial injury response and by low oxygen. A trajectory was reconstructed from 23 of 58 subclones and seen in six of nine tumours. Direction along the path is inferred from an RNA-velocity model, but an organoid test did follow time: one development-like line gave rise to injury and hypoxia cells. The twelve tumours are all one genetic subtype, and the clinical suggestions, targeting the path and the myeloid environment around it, are proposals, not tested treatments. Frontier entry · Paper

Disruption of cell-type-specific molecular programs of striatal spiny projection neurons in autism

Frontier Score 66 · Neuroscience · Cell

Autism research has centred on the cortex; this work points deeper. Single-nucleus multiomics on the putamen of 84 donors (47 with autism, 37 neurotypical) found the disrupted genes concentrated in one cell type, D1 striosome spiny projection neurons, hitting synapse and energy-metabolism genes, and flagged two regulators, EGR1 and EGR3, whose deletion in human striatal organoids reproduced part of the shift. The human data are associational, from 346,556 profiled nuclei across a wide age range; the causal arrows come from a mouse carrying one specific autism-linked deletion and from the organoids, which establish the regulators without reproducing the disorder. Frontier entry · Paper

Robust fine-mapping in the presence of linkage disequilibrium mismatch

Frontier Score 70 · Biomedicine & Health · Nature Genetics

Fine-mapping uses GWAS summary statistics plus the correlation between nearby variants, usually borrowed from a reference panel rather than the study's own participants, and a mismatch is common and degrades accuracy. RSparsePro is built to tolerate it, and in simulations and in fine-mapping protein and complex-trait loci its credible sets showed higher power, coverage and replication than existing methods. The variants it finds are statistical candidates, and the regulatory mechanisms it implicates are, in the authors' own word, potential. Frontier entry · Paper

4. The brain, and a quantum limit

An astrocyte-microglia MFG-E8 axis promotes synapse engulfment in early amyloid pathology

Frontier Score 70 · Neuroscience · Neuron

Losing synapses in particular regions is an early hallmark of Alzheimer's, but why those synapses are vulnerable is unclear. In a mouse model of amyloidosis, the authors identify an early astrocyte state in vulnerable regions, with bulbous processes and impaired autophagy, that appears before widespread plaque. Rather than eating synapses themselves, these astrocytes raise MFG-E8 and sit beside increased synapse engulfment by microglia; switching MFG-E8 up and down shows it is required for that engulfment, and deleting it from astrocytes reduces synapse loss across models. It is a mouse result about a signalling step, not a treatment, and much of the detail comes from the open preprint, where the engulfment counts rest on 4 to 6 mice per condition; the cognitive benefit is stated only in the paywalled abstract. Frontier entry · Paper

Information formatting in neuronal populations predicts differential functional contributions to behavior

Frontier Score 68 · Neuroscience · Nature Neuroscience

Why do distinct brain areas exist if neurons everywhere encode almost everything? The authors propose that areas differ in how they format information. Two monkeys reported motion direction over 41,106 choices while the authors recorded from the motion area MT and the prefrontal cortex: both encoded motion and reward, but MT kept them separable while the prefrontal cortex combined them in a decision-related format. A network model predicted that stimulating the two areas would affect choices differently, and microstimulation agreed, MT biasing choices toward an average and the prefrontal cortex producing winner-take-all outcomes. The study rests on two monkeys and a deliberately simplified model used as a hypothesis generator. Frontier entry · Paper

Optimal learning of quantum channels in diamond distance

Frontier Score 70 · Quantum · Nature Physics

Process tomography, learning an unknown quantum channel by probing it, underpins the calibration of every quantum computer, and how many uses it takes in the diamond distance, the strict worst-case measure, had been an open question. The authors settle it: a channel can be learned to fixed accuracy using a number of uses that scales with the parameters it actually has, with a matching lower bound showing no method can do better. For a general channel on a d-dimensional system that is d^4 uses, down from the d^6 of previous schemes, and because channels include states, gates and measurements, one framework covers them all. The bound is a worst-case asymptotic scaling, not an exact count for a small device. Frontier entry · Paper

5. Cross-checks and bounds

Three of this week's results do not report a new observation but test an existing one, or close off an explanation. That work rarely makes a headline and is exactly what a young field needs.

Independent interferometric observable confirms black hole shadow in M87*

Frontier Score 62 · Space & Astronomy · Nature

The 2019 Event Horizon Telescope image of M87* left the ring shape sensitive to processing choices, and some reanalyses did not recover a clear ring. The authors reconstructed the image with an algorithm built on closure invariants, quantities immune to the calibration errors that plague this interferometry, and with no built-in assumption about the source, and across three observing years it unequivocally reproduced a ring, at 42.2 ± 1.5 microarcseconds, implying a mass of 6.5 ± 0.5 billion Suns. It re-analyses the existing data rather than observing anew, so it is a cross-check; its strength is that the closure-invariant approach is insensitive to the choices that drove earlier disagreements. Frontier entry · Paper

Exorcising ghosts with gravitational waves

Frontier Score 66 · Space & Astronomy · Physical Review Letters

Adding curvature-squared terms to general relativity, a common way to probe short-distance physics, generically brings in a massive spin-2 "ghost" that makes the theory non-unitary. The authors compute the gravitational waves both the ghostful and a ghost-free version emit and confront them with data: in the ghostful theory, when the extra modes are light the quadrupole formula of general relativity is not recovered, conflicting with binary-star measurements, and using pulsar binaries and GW170817 they set a lower bound of 4.6 × 10⁻¹² eV on the extra modes' mass. It is theory confronted with existing data, worked in the linearised regime, and the authors call it the first gravitational-wave test of fourth-order gravity. Frontier entry · Paper

Modeling direct waves in binary black hole ringdowns

Frontier Score 62 · Space & Astronomy · Physical Review Letters

It had been suggested that the "direct wave" in a black hole merger signal, the prompt part travelling straight to the observer, is tied to the event horizon and could probe its geometry. Testing the idea on numerical-relativity waveforms, the authors identify the direct wave across a range of start times but find its frequency deviates from the horizon value, drifting 10 to 20% over the ringdown, which limits its use as a horizon probe. A clean negative result on simulated waveforms, not detections. Frontier entry · Paper

6. Medicine: a definition, a prevention trial, and five therapies

Validation of clinical and preclinical obesity across global cohorts

Frontier Score 65 · Biomedicine & Health · Nature

A 2025 Lancet Commission proposed splitting obesity into clinical (excess fat plus organ dysfunction) and preclinical (excess fat, preserved function), adding waist measures to BMI. Tested against BMI alone in 844,108 adults across four countries, the combined definition changed who counted as obese and, crucially, tracked mortality where BMI-defined obesity alone did not: clinical obesity carried a hazard ratio of 1.69, preclinical only 1.19. Organ dysfunction was read from diagnostic codes, not standardised exams, and the clinical-obesity prevalence figures are attributable-fraction estimates, not raw counts. Frontier entry · Paper

Prevention of Venous Thromboembolism after Total Hip or Knee Replacement

Frontier Score 66 · Biomedicine & Health · NEJM

In a noninferiority trial that randomized 18,883 patients after hip or knee replacement to aspirin, warfarin or rivaroxaban, aspirin could not be shown to prevent symptomatic clots and deaths as well as the two anticoagulants, which in turn could not be shown to be as safe. Not noninferior is not the same as inferior: each effectiveness interval also included no difference, so the trial could not rule out a disadvantage for aspirin rather than showing one. Every event rate was low, a few per hundred patients or fewer. Frontier entry · Paper

Five therapy results round out the week; three are randomized trials and two are single-arm, with no comparison group.

7. Where the points went

The week's most-cited paper is not its top-scored one, and that is the design working as intended. BeyondMimic (68), a humanoid-control method that composes agile skills learned from human motion, has by far the week's highest Impact pillar, 92.1: published in late August, it has already gathered 315 citations and a field-weighted citation impact of 561. Yet it lands at 68, below the Higgsino paper's 76, for two reasons the score makes explicit.

First, Novelty is weighted at half the score, and BeyondMimic's is ordinary (63.4) for an engineering advance built on a well-worked pairing of robot-learning topics. Second, the recency factor: at three months old, BeyondMimic keeps 89.6% of its pillar-weighted standing, while the Higgsino paper, three days old, keeps 99.6%. That gap, roughly ten points of weighted score, is what separates them. The highest novelty of the week belongs instead to a paper near the bottom of this chart: a cancer-selective gene circuit (66) keyed to a quirk of how tumour cells misread their own genetic code, on a rare topic pairing (161 prior works), whose Novelty reads 77.1 but whose Impact is only 30.3 because it has barely been cited and is closed access.

The score deliberately favours results that are new, well established and starting to be cited over results that are merely popular, which is why the pillars are published next to every entry.

8. Also new this week

The remaining 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

  • Cancer-selective gene circuits via intrinsic ribosomal frameshifting (66, Nature). Specific codon pairs trigger a shift in the ribosome's reading frame preferentially in cancer cells, and the authors use it to build a synthetic circuit that makes diverse cancer cells kill themselves while sparing normal tissue, with tumour regression in two mouse cancer models. Written from the abstract alone, so the numbers behind each claim are not quotable, and the normal-tissue safety is an absence of observed effect. Paper
  • Granzyme K is a driver of age-related tissue remodeling (63, Immunity). In mice bred to lack the immune enzyme granzyme K, immune ageing persisted but age-related decline in the kidney and liver eased, uncoupling the two. A mouse genetics result about specific organs and markers, not a drug and not lifespan. Paper
  • p53 activates chaperone-mediated autophagy to suppress tumorigenesis (61, Developmental Cell). The tumour suppressor p53 is found to switch on a selective protein-disposal pathway that degrades two metabolic enzymes and curbs a metabolic route tumours lean on, slowing lung and liver cancers in mouse models. A biochemical chain demonstrated in model systems, not patients. Paper
  • TIGIT and PD-L1 co-blockade promotes dendritic cell-dependent CD8+ T cells (59, Immunity). The mechanism behind a two-antibody cancer immunotherapy turns out to run through dendritic cells, not the T cells directly, with type I interferon sustaining the loop, established causally in mouse tumours. Paper
  • Early assessment of the Bundibugyo virus outbreak 2026 (61, Science). Combining case records with mobility data, the authors find human movement a principal driver of the early geographic spread of a 2026 ebolavirus outbreak in the DR Congo, with an early doubling time of about 10 days. A description of how it spread, not why it started. Paper
  • Long-term outcomes after PCI versus CABG for left main disease (61, the Lancet). In a registry of 5,379 Asian patients with left main coronary disease followed a median 10 years, stents and bypass surgery showed no statistically detectable difference in mortality (24.0% against 22.9%), which the authors stress is not evidence of equivalence. Paper

Physics

  • Atomically resolved superconducting gap in (La,Pr)3Ni2O7 films (61, Physical Review X). Scanning tunnelling spectroscopy on bilayer nickelate films gives reproducible U-shaped spectra with a nearly zero-conductance bottom when transferred quickly in cryogenic vacuum, which the authors read as a fully opened superconducting gap in oxygen-sufficient films. Paper

AI & Computing

  • Foundation model for particle collision data analysis (64, Nature Machine Intelligence). EVENET, pretrained on 500 million simulated collisions, builds one shared representation and fine-tunes per task, beating baselines especially when training data are scarce and rediscovering the Upsilon meson in CMS Open Data at 7.6 sigma. Apart from that rediscovery, every test is on simulation. Paper
  • BeyondMimic: versatile humanoid control via guided diffusion (68, Science Robotics). A compact motion-tracking formulation masters agile skills (cartwheels, spin kicks, sprinting) with one setup, and a latent diffusion model composes and redirects them at run time, transferring to a real humanoid with no further training. Trained entirely in simulation; what ran on hardware is a subset. Paper

Quantum

  • Observation of prethermal dynamical phase transitions and resonant melting (67, Nature). A neutral-atom simulator running up to 180 qubits maps long-lived prethermal dynamical phases and the resonances that destroy them, including a first-order dynamical transition in two dimensions, in a regime hard to compute. The states are long-lived, not permanent, and the two-dimensional transition likely does not persist at long times. Paper
  • Extensive mixed-state entanglement in kinetically constrained superradiance (59, Physical Review Letters). Theory shows that a local constraint between emitters lets a superradiant burst leave them extensively entangled while keeping superradiance's signatures. A proposal supported by small-system simulations, with the entanglement transient under single-site loss. Paper
  • Controlling Rydberg atom-polariton interactions (58, Physical Review X). An experiment maps three regimes of interaction between a Rydberg polariton and an adjacent Rydberg atom, passing through an exceptional point, and tests the setup for fast, non-destructive readout, with a fidelity the authors say is not yet sufficient for use. Paper
  • Probing the memory of a superconducting qubit environment (56, Physical Review Letters). Non-Poissonian quantum jump traces are shown to carry the information needed to distinguish long-lived defects, which retain memory of a qubit's past states, from a memoryless bath, pinpointing several such defects by sweeping the qubit's frequency. Indirect evidence on one qubit, with simple modelling. Paper

Materials & Energy

  • Scale-rich network-based metamaterials (68, Nature). A framework grows strut networks whose lengths, thicknesses and connectivity span orders of magnitude from two parameters, covering 77% of a density-anisotropy map and, in 3D-printed samples, deforming without the compaction bands that form in lattices. The experiments are on planar, centimetre-scale polymer samples. Paper
  • Twist-angle-dependent quantum phase diagrams in moiré bilayer MoTe2 (63, Nature). A transport study across 3.79° to 5.76° of twist maps an evolution from fractional topological states to correlated states and superconductivity above about 5.6°. Transport is an indirect probe, and the origin of the superconductivity is left open. Paper

Climate & Environment

  • Increased demographic turnover and isolation precede social fragmentation in Asian elephants (67, Science). Tracking 443 wild female elephants over 12 years, the authors watched the social network fracture into smaller, isolated groups as deaths and departures outpaced arrivals, with a core group present every year fragmenting too after a lag. Observational, at one site, and the timing is correlational. Paper
  • Human land use drives microplastic abundance in terrestrial wildlife (62, Nature). Blood from 26 tapirs and 21 anteaters in Brazil, paired with GPS tracking, found microplastics in nearly all of them and higher levels the closer animals lived to development and waste sites, though also in every animal more than 65 km from the nearest development. An observational field study; the land-use links are associations across 40 animals. Paper

9. What the week says

A field finding its feet. Three altermagnet papers landed in one week, and read together they show how a newly named phase matures: direct evidence for its predicted chiral spin wave, a device that injects spin without net magnetization, and signatures surviving in films four unit cells thick. None is a closed case, and each says so, calling its material a candidate or its order a set of signatures. This is what the early years of a phase look like, several groups converging on the same claim by different measurements, none yet decisive alone.

Cross-checks over discoveries. Three of the week's physics and astronomy results test an existing claim rather than make a new one: an independent reconstruction confirms M87*'s shadow, a horizon-probe idea for black hole ringdowns is found too imprecise, and gravitational waves bound the extra modes of a modified gravity. None is a headline, and all three are the part of science that keeps a result honest.

Where a gene's effect shows up. As in recent weeks, several teams spent their effort locating where a gene or variant acts: tandem repeats mapped to immune cell types, autism risk converging on one striatal neuron class, glioblastoma states resolved into a single path. All stop at association or at model systems, and all say so.

An honest answer on AI for science. The week's second-ranked paper asked whether AI can generate research ideas scientists find interesting and reported, carefully, that mostly it cannot yet, and that the knowledge-graph machinery meant to help did not beat a plain prompt. A negative result on a question many are hyping, run as a real expert evaluation, is worth more than another demo.

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.

Correction, 18 November 2026: a refresh after this issue went out re-read the Impact of the glioblastoma cell-trajectory atlas and moved its score from 70 to 68. Its score line and the chart above now show the new value.

The Frontier Brief

The week's frontier, scored.

Every Monday, the highest-scoring new breakthroughs in the index, each with its score and a one-line read on why it matters. What you get

One email a week. The top new breakthroughs, scored. Unsubscribe anytime.