Same words, different measurements, and a ward score that missed two points

A chemical-biology paper names a failure mode of structure prediction and gives it a word: models place ligands where the training data already were. Two other fields published the same complaint about units on the same day, one for nuclear mechanics, one for sleep trackers. In an English assessment unit, a deterioration score was computed without the oxygen a patient was receiving and without the confusion she had developed. A nanopore workflow in Sheffield put a price on the small laboratory: cost-neutral at nine samples per run. Two randomised trials reported no effect, and one of them measured why.

Connecting the dots

Three fields discovered the same day that their numbers cannot be pooled. "Towards a rational terminology for cell types" argues that cell-type labels have no operational definition across studies or species. "Nuclear mechanomics beyond stiffness: toward a genuine mechanical state space" says the instruments that measure a nucleus are not interconvertible. "Standardizing wearable sleep data for clinical, research, and public health use: A World Sleep Society specification" says manufacturers report measures with the same names and different definitions. Each is the cross-instrument generalisation problem logged on 2026-10-13, arriving in three unrelated literatures. The sleep group is the only one with a shipped answer, and it is to standardise the back end and leave the sensors alone.

A sum with two missing inputs decides when antibiotics arrive. England's safety investigator found a patient's oxygen requirement documented but left out of her early warning score, and her new confusion never scored at all. Under the national framework those two omissions move her from a six-hour antibiotic window to a three-hour one. The measurement-based care trial published today is the same break one layer up, since therapists discussed the questionnaire feedback in 29.3% of cases and masked symptom ratings moved not at all. It is the stalled human link of the electrocardiogram screening trial of 2026-10-17.

Models are confident where the data already were. The co-folding result below is quantified bias from imbalanced training. The binding-affinity benchmark below is the mirror case, where free text describing an assay carries predictive signal that standard curation discards. Speculation: both point at the data-selection procedure as the thing worth auditing rather than the score, which is Moritz Hardt's protocol argument of 2026-10-17 applied to chemistry.

New concepts

"Orthostery burnout": a model predicts pockets where its training set lived

"Most stereoprobe liganding events occurred at nonorthosteric sites and evaded assignment" — Hayward et al., Nature Chemical Biology

"Boltz-2 rarely predicted the location of the much larger subset of ligand interactions" … "which we term 'orthostery burnout' to reflect bias likely induced by imbalanced training" — same

The asymmetry is extreme. Near a known orthosteric pocket the model placed the ligand correctly in "239 of 252 predictions"; away from such pockets, "22 of 548 predictions". Across the set, "Only 23 of 102 proteins" produced at least one structure with the ligand within 5 Å. The reason it matters to the authors is stated plainly: "Most proteins in the human proteome lack chemical probes." New pharmacology is found outside well-characterised pockets, which is where the prediction fails.

Figure 2, Hayward et al., Nature Chemical Biology 2026, CC BY 4.0, link. Published title: "Boltz-2 analysis of stereoprobe-liganded proteins."

Name a cell type after the moment in evolution it first appeared

"Instead, we propose introducing a phylogenetic prefix that reflects evolutionary origins." — Zhu et al.

The diagnosis is that "these labels lack consistent, operational definitions across studies, organs, and species" and "no universal criteria exist to determine when two cell populations should be considered as distinct". The proposal puts a prefix at the deepest clade the evidence supports, giving "Vertebrata astrocytes and Vertebrata ependymal cells" or "Assigning a Cnidaria prefix to cnidocytes", while "we are not trying to abandon current cell-type names." Every name then carries a falsifiable phylogenetic claim, which is a different object from an ontology label.

Nuclear mechanics has no state space, so physicists propose one

"Mechanical measurements of the nucleus remain fragmented across modalities" and "are often reduced to scalar "stiffness" proxies" — Herráez-Aguilar et al., npj Systems Biology and Applications

They define a "regime-aware, physically interpretable representation of intrinsic nuclear mechanical state" and name what building it needs: "state preservation, identifiability, uncertainty propagation and cross-regime validation". No dimensionality is claimed. The useful part is negative, since "Readouts from non-equivalent techniques are converted into model-dependent descriptors", so mechanics cannot be appended to an atlas as one more axis.

Basic biology

The marker for human totipotency marks an intermediate, and a retrovirus does the work

"the previously utilized TPRX1 reporter marks an intermediate population" which "subsequently transitions into HERVL-positive 8CLCs" … "knockdown of HERVL impaired 8CLC generation", indicating "a cis-acting role for HERVL in activating the 8CLC transcriptional program" — GEO series summary, GSE309040

The readout is a "HERVL-tdTomato/TPRX1-GFP dual-reporter hPSC line", so the two states can be sorted apart. An endogenous retrovirus turns out to be an actor in the program rather than a label on it.

Six commodity plastics make reactive oxygen species in plain water

"Here we show that commodity microplastics exhibit intrinsic physicochemical activity in water" with "spontaneous generation of reactive oxygen species that increases with interfacial hydrophobicity" — Qian et al., Nature 658, 650–657

The activity "strongly correlates with interfacial electrostatics", and "hydroxide oxidation becomes energetically more favourable at increasingly hydrophobic microplastic–water interfaces". The authors say this has "direct implications for biomolecular adsorption, protein aggregation, and even cellular damage". Six polymers were tested, from polypropylene to polymethyl methacrylate. The tissue-exposure studies of 2026-10-13 and 2026-10-14 counted particles without a mechanism, and the active variable here is surface chemistry, which a particle count does not record.

Atlases and datasets

• A cysteine ligandability map. "Across >24,000 cysteines quantified in human cancer cells", with mass spectrometry in PRIDE as PXD069511 and code at github.com/cravattlab/Hayward_2025_Boltz_analysis. — Nature Chemical Biology
• A benchmark that keeps the assay text. Conditioned on the free-text assay description, "MAMMAL-AsAP achieves the highest (R^2) of 0.7198 +/- 0.0085", because "assay descriptions contain predictive information not fully captured by molecular structures and protein sequences". — Gupta et al., bioRxiv

Methods

Measuring what a mutation does to activity, separately from how much protein there is

"fitness scores comprise a conflation of expression level (i.e. enzyme abundance) and catalytic activity" — Vanella et al., Nature Communications 15, 1807, CC BY 4.0

Enzyme proximity sequencing turns catalysis into a fluorescent mark with a "horseradish peroxidase (HRP)-mediated phenoxyl radical coupling reaction at the yeast surface", then sorts and sequences, giving "catalytic activity phenotypes of thousands of variant enzymes from a cellular pool in a single experiment". The published scan covers how "6399 missense mutations influence folding stability and catalytic activity". Every scan that reports one number per variant has been adding those two quantities together.

Sorting cells that were touching from cells that shared a droplet

"integrated spectral imaging cell sorting with physical cell deconvolution" to separate "physically interacting (in-contact) and non-interacting (coincidental) populations" — Kang et al., bioRxiv, CC BY 4.0

In T cell and monocyte doublets from tuberculosis patients the group reports "unidirectional monocyte-to-T cell RNA transfer", with coincidental doublets sitting between singlets and true contacts, read as "recently disrupted or forming physical pairs". The ambiguity this sort removes has sat under every previous experiment on interacting cells.

Also new: a force-field fit posed as "a bilevel optimization problem" solved "using the implicit function theorem", which takes the gradient through the converged structure and leaves an existing energy minimiser in place (Tamby, Bujupaj and Orlando, bioRxiv).

Clinical trials

Feeding symptom questionnaires back to therapists changed nothing a blinded rater could see

"masked ratings of total symptoms (SFSS-I) showed no effect" … "but its efficacy remained insufficient, likely reflecting implementation challenges" — Hautmann et al., Psychotherapy Research, DRKS00016737

"317 children (6-10) and youth (11-17) were randomized to behavioral therapy" in routine care. Completion of the questionnaires was high; use was not, since "therapists discussed the report at least once in only 29.3% of cases". Unblinded caregiver ratings showed small improvements.

Dapagliflozin in heart failure patients selected for having no fibrosis

"dapagliflozin did not improve LV distensibility or relaxation" — Scheffer et al., European Heart Journal Cardiovascular Pharmacotherapy

A "34-week, randomized, double-blind, placebo-controlled crossover trial" where "32 patients were randomized and 29 completed the study protocol", with co-primary stiffness endpoints at "p=0.370" and "p=0.448". Patients were chosen by imaging phenotype rather than diagnosis, and the mechanism still did not move.

Hospitals

A fire cut 3.46 million emergency-room and bed records out of exchange

(translated from Korean) "the data that could not be linked came to 7 types, 3,544,973 records in total", of which "emergency room and bed information, amounting to 3,459,171 records" was the largest share. — Korean Doctors' Weekly, 16 Oct Original: "연계되지 못한 데이터는 7종, 총 354만4,973건이었다" … "응급병실·병상 정보로, 345만9,171건에 달했다"

For 35 days nothing new was reflected, and some links were resynchronised "about 10 months after the fire". No attacker was involved, and what went missing was the shared view of which beds were free. Compare the vendor breaches in Japan and the regional outage in Castilla y León of 2026-10-16. The same payer is also heading for an operating deficit of 6.9161 trillion won this year, with reserves "expected to be exhausted around 2029" (17 Oct; original: "이 추세라면 적립금이 2029년께 소진될 것으로 예상된다").

A sepsis death where confusion was never scored

"Lorna was more unwell on her admission to the AAU than staff realised." … "New confusion in patients is not consistently accounted for in NEWS2 scores." … "Staff perceive that a diagnosis of infection is needed before completing the sepsis screening tool." — HSSIB, third investigation report

"Her need for oxygen was documented but not included in the scoring.", and "If the need for oxygen (which scores 2) had been included, her NEWS2 score would have totalled 6." Later her recorded 9 "did not include the score for confusion, which if added, would have resulted in a NEWS2 score of 12." The report adds that "Nursing staff were reluctant to answer the first question of the sepsis screen".

The population version: in "29,647 patients (median age, 71 years)" across 67 hospitals, "gastrointestinal symptoms were associated with decreased odds of timely antibiotics" while "prehospital fever was associated increased odds", and timely antibiotics came with 30-day mortality of "18.9% vs 21.0%" (CIDRAP, on a JAMA Network Open study).

Medicine concept of the day

NEWS2, the score that decides who gets seen and how fast antibiotics come

It is arithmetic on observations a nurse already takes. "Six simple physiological parameters form the basis of the scoring system", namely respiration rate, oxygen saturation, systolic blood pressure, pulse rate, temperature and "level of consciousness or new confusion". "A score is allocated to each parameter as they are measured", and then "The score is then aggregated and uplifted by 2 points for people requiring supplemental oxygen". New confusion means "The patient has new-onset confusion, disorientation and/or agitation, where previously their mental state was normal" and "scores 3 on the NEWS system". It "has received formal endorsement from NHS England and NHS Improvement" (Royal College of Physicians).

Since 2022 the number also starts the antibiotic clock. Scotland's guideline summarises a framework that "provides a severity score based on NEWS2 bands of 0, 1–4, 5–6 and ≥7", with antibiotics expected within "1 hour of recording a NEWS2 score ≥7 for patients with definite infection", "3 hours of recording a NEWS2 score of 5–6 for patients with probable infection" and "6 hours of recording a NEWS2 score of 1–4 for patients with possible infection", timed from "the time of the first NEWS2 assessment on presentation to the emergency department" (SIGN 167 toolkit). The two unscored points in the case above were therefore not bookkeeping.

The weakest input is the one that failed in that ward. Against a validated delirium instrument in "13,908 consecutive non-elective hospital admissions", "Sensitivity of NEWS2 for 4AT ≥4 was 17.8%" while "specificity was 99.1% (95% CI 98.9-99.3)", so "NEWS2 in current practice has low sensitivity but high specificity for delirium" (Vardy et al., Clinical Medicine 2022). A model trained to predict this score inherits that item.

Worth knowing

Nanopore 16S sequencing, with the operational numbers attached. The workflow "significantly expanded clinical diagnostic yield from 14.0% (12/86) to 37.2% (32/86) (p<0.0001)" on culture-negative samples and "identified polymicrobial infections (n=8)", with "a 100% positive percent agreement (12/12; 95% CI: 75.7%-100.0%)" against outsourced testing. Two claims decide whether a district hospital can run it, since it "operated without specialist bioinformaticians" and reached cost-neutrality "at 9 samples per run", below which it costs more than sending samples away (Saville et al., Journal of Infection).

Kenya's Ebola count went from three to seven. The ministry confirmed "four additional cases of Ebola Virus Disease (Bundibugyo virus)", "bringing the total number of confirmed cases in the country to seven, including one death", and "They include three healthcare workers and one family member." Containment is holding at tracing, since "all four new cases were detected among individuals already under quarantine", "128 of the 130 identified contacts (98.5%) have been traced", and there is "no evidence of community transmission at this stage" (Africa CDC, 17 Oct). No vaccine has been shown to work against this species (2026-10-14).

Bangladesh has counted 1,177 measles-associated deaths since March and confirmed 104. Since 15 March "a total of 1,177 people have died from measles or symptoms associated with the disease", with "104 among confirmed measles cases" and "1,073 deaths among people with suspected symptoms", while "the number of people presenting at hospitals with measles symptoms has reached 208,570" (bdnews24, 14 Oct). Nine in ten of those deaths were never laboratory-confirmed, by my arithmetic on the article's own figures.

Central Africa, 250 tonnes of seized medical products in five years. (translated from French) "More than 250 tonnes of medical products were seized" "in the six CEMAC countries between 2020 and 2025", in a new report by the United Nations Office on Drugs and Crime, and "Recorded seizures went from about 5 tonnes in 2020 to 116 tonnes in 2024." (allAfrica, 17 Oct; originals: « Plus de 250 tonnes de produits médicaux ont été saisies », « dans les six pays de la Cémac entre 2020 et 2025 », « Les saisies recensées sont passées d'environ 5 tonnes en 2020 à 116 tonnes en 2024. »). Seizures count what customs caught.

Health systems and policy

A sleep-data standard that leaves the hardware alone

"Although many manufacturers report similarly named sleep measures" … the group proposes "a minimum viable standard, specifying a common semantic framework for Fundamental Sleep Measures" — Chee et al., Sleep Medicine

The reason is downstream: "Population reference values, risk thresholds, and outcome associations can only be established reliably" once measures mean the same thing, and the method is "standardizing backend data semantics rather than frontend product implementation". The society's release states that "The specification does not prescribe which sensors should be used" and adds "a reproducible "Sleep Percentage" based on sleep occurring between physiological sleep onset and offset" for devices that cannot tell whether the wearer was in bed (World Sleep Society, 13 Oct). Any consumer device can comply, because the burden is on what it exports.

Device approval times differ eightfold across fourteen countries

"Approval timelines ranged from 3 to 24 months." … "Despite harmonization efforts, significant variations in regulatory standards persist across countries." — Chongmelaxme, Saharan and Kua, Value in Health Regional Issues

The survey covers the twelve member countries of the International Medical Device Regulators Forum plus Pakistan and Thailand, with "12 countries using a 4-tier model and 2 using a 3-tier classification" and "ISO 13485 certification was widely required for quality management system". The technical requirements are largely shared across the fourteen, and the waiting time is where they differ by a factor of eight.

Where the reference data are missing, said by someone working there

"Community studies demonstrate substantial diagnostic invisibility" and "unequal access to neurological diagnosis" — Kipanyula, Neuroscience

The genetic example is concrete, since "The glucosylceramidase beta 1 (GBA1) rs3115534-G Parkinson's disease signal provides a proof of principle" of "mechanisms that are poorly captured in ancestry-restricted datasets", from which "thresholds and models require population-specific validation". A biomarker cut-off is a local quantity that has so far been borrowed.

Patents

Merck has claimed pure water as the primary calibration standard for high-sensitivity optical spectrometers. "A high-purity H2O solution is used as the primary standard, by way of measuring Raman scatter.", and the one non-obvious step is "Applying a correction methodology for correction of variations in the Raman scattering caused by the container". The stated problem is price, since "the Raman standards are usually expensive (SRM 2241 >$3500)" (WO 2025/119804). Instrument-to-instrument optical efficiency is what makes spectra comparable across sites, and the cheapest reference for it is now claimed procedure.

Startups and deals

A recurrence score read off the breast MRI the patient already had. SimBioSys received "510(k) clearance to TumorSight Risk", "an artificial intelligence software device that analyzes breast MRI scans" that "requires no additional procedure, tissue shipment or specialized hardware", producing "a proprietary score from zero to 100 and a categorical high-, intermediate- or low-risk classification" for "adult women with hormone receptor-positive, HER2-negative Stage I–IIIA breast cancer". Validation was "a multi-institutional study of 2,129 patients" from "Mayo Clinic, Emory University, Stanford Medicine and Mercy Medical Cedar Rapids" (Unite.AI, 15 Oct). A prognostic answer that usually needs a tissue assay comes from an image already acquired.

Radiology consolidating into one stack. Sirona Medical raised "$40 million in Series D financing led by Avidity Partners" for a platform that "unifies PACS, reporting, diagnostic viewing, worklist, and workflow AI on a single cloud and AI-native platform", plus "Cortex, Sirona's agentic AI orchestration layer" (BioSpace, 15 Oct).

Repos and tools

All of arXiv as files, once. A snapshot that "covers 3,148,796 papers and includes file contents, paths, sizes and SHA-256 digests", where "All nine configs total 16,076,056,758,128 bytes, or 16.08 TB." The card is careful: "It is a one-off snapshot; coverage gaps are documented below.", and "The compilation's CC0 dedication does not apply to the papers" (Hugging Face).

A synthetic clinical corpus that audited its own citations. Consultations "regenerated from scratch with Claude Opus 5.5, one agent per disease", covering "2,194 clinical encounters covering 2,194 unique human diseases" with "6,377 verified PubMed references across the records (4,198 distinct PMIDs)". The finding worth keeping concerns the inherited material: "the curated source lists turned out to be mostly wrong: only 5.8% of their PMIDs point at the paper their title claims" (Hugging Face, Apache-2.0). Anyone building a literature graph on a hand-curated reference list should check that number first.

AI: using it well, what's new

Four agents, one Minecraft world, twenty-seven hours. "We placed four Gemini 4 Argon agents in a sandboxed Minecraft survival world" and "No one assigned roles or wrote game logic."; "the team wrote 2,783 programs and defeated the dragon in 27 hours and 14 minutes, dying 605 times along the way." The paper marks its own limits: "Mineflayer tells the agents exactly which blocks and creatures are around them", "hunger cannot kill a player.", and "We describe a single run, the first launch we tried, which cannot show how reliably the setup succeeds." (Faldor, Singh and Raileanu, arXiv, CC BY 4.0). The action space is programs rather than keystrokes, the same abstraction as the laboratory stack of 2026-10-17.

Figure 1, Faldor et al., arXiv 2026, CC BY 4.0, link. Published caption: "The four agents, and their run step by step: the thirteen steps from bare hands to the Ender Dragon," "each colored by the player who reached it first, over 27 hours and 14 minutes."

An ordering of the frontier, and a licence clause that decides what a laboratory may use. Nathan Lambert writes that "This has been a week where the balance of American open-weight models took a step change back toward the frontier." and "The ranking would be Anthropic > OpenAI > Meta ~ Gemini > xAI ~ Chinese Open Weights > U.S. Open Weights." The catch sits in the terms, since "Users of the models in products must display "Built with Muse" on their products.", while "The license does permit using Muse Spark Open for internal research and applications." (Interconnects, 17 Oct). Two days later DeepSeek posted a checkpoint that "reduces task execution time and output length" and claims to be "advancing the performance–cost Pareto frontier across multiple agent harnesses", with Terminal-Bench 4.0 going from 31.2 to 42.3, under MIT for weights and code (model card).

Other frontiers

The expensive half of fault-tolerant quantum computing got cheaper. Google reports that "we've shown that "magic state cultivation" works in practice on our Willow processor", against a baseline where "creating high-quality, error-corrected magic states is incredibly resource-heavy." and "up to 95% of a quantum computer's entire capacity might need to be dedicated to making them". The result is "a 40-fold reduction in errors relative to un-cultivated states" and "T-states with a record logical fidelity of 99.99%", with the caveat that "we still need to demonstrate magic state cultivation at a larger scale" (Rosenfeld and Gidney, Google Quantum AI, 15 Oct; paper in Science).

Corners of the web

Back from Khartoum, two doctors from an Italian non-governmental organisation describe the first job as looking. (translated from Italian) "But faced with a city like Khartoum the first task should be simpler: to see, listen, understand." A maternity hospital that handled "about 3,700-3,800 births a year" is still closed, "only about forty of the roughly 120 hospitals of Khartoum" put out of service have resumed, and "There is still no reliable estimate of how many health professionals are missing today in the capital" (Putoto and Maretti, Medici con l'Africa CUAMM, 12 Oct; originals: « Ma davanti a una città come Khartoum il primo compito dovrebbe essere più semplice: vedere, ascoltare, capire. », « circa 3.700-3.800 parti all'anno », « soltanto una quarantina dei circa 120 ospedali di Khartoum resi non operativi dalla guerra », « Non esiste ancora una stima affidabile di quanti professionisti sanitari manchino oggi nella capitale »).

People and good examples

Lead chromate in turmeric, from isotopes to a televised enforcement day. In Bangladesh "lead chromate had been added to turmeric for four decades" and a policy existed without enforcement; afterwards "Within a few months, there was a 90% reduction in the prevalence of lead-tainted turmeric", and "less than US$ 1 million was spent" (Forsyth et al., npj Science of Food 10:220, CC BY 4.0). The mechanics: "It was a handheld X-ray fluorescent analyzer.", bought "for the team for roughly $40,000"; "The government declared that turmeric adulteration was a prosecutable violation"; a sting was "filmed by a Jamuna Television news crew and broadcast across the nation."; adulteration "plunged from 47% in September 2019" to none detectable in 2021; and "16 months after the intervention, lead levels in the blood of sample test subjects dropped by a median of 30%." Jenna Forsyth's description: "It was law enforcement theater." (Newby, Stanford Medicine). In eastern India "30% of the turmeric samples (n = 153) across 25 cities had lead levels above 10 µg g⁻¹", and the barrier named is "sample lab testing to confirm lead chromate adulteration".

Thirty-five years of sending medics instead of police, ended by its funding. In Eugene, Oregon, "CAHOOTS responds to nearly one-fifth of emergency calls that would have otherwise been addressed by EPD", and the authors' re-estimate corrects the operator's own number, since "We estimate the SDR at 18%" against "the previous EPD estimates of 3 to 8%". The service "has the ability to transport clients and provide basic medical care", and "CAHOOTS ceased services in Eugene in April 2025 because of a funding problem in the nonprofit that runs the program" (Burton, Herbert and Rohlfs, Science Advances 12(39), CC BY-NC 4.0). Dispatch logs and a change of operating hours gave a cheap evaluation design for a service copied in more than a hundred cities.

Reads and ideas

Training as a by-product of real work. Ben Reinhardt: "Serious technical training needs to happen while doing serious work." The diagnosis is blunt, that "The quality of elite technical education in the US is abysmal." and "In 2026, you can graduate from an Ivy League school with a technical degree" without linear algebra, while "work portfolios are starting to matter far more than credentials." His conclusion is that "the way to forge hardcore scientists and technologists is not to build yet another school.", and the vehicle he proposes is an industrial research laboratory taking apprentices, with "no grades, no degree, and no accreditation" (Speculative Technologies).

What $60 million of research-reform grants bought. Stuart Buck: "The grants I made totaled over $60 million dollars." The cheapest did the most, since "At $250k, it was easily the highest impact grant I ever made." and "Only around 40% (give or take) of the 100 experiments could definitively be replicated". The failures are the instructive part, since of one multi-group project "the grant stalled and never produced anything that was very functional.", and of a trial-data platform, "But not enough people request Vivli data." (The Good Science Project). A repository with no demand and a consortium with no owner are two ways to fund infrastructure and get nothing.

Book of the day

Fevers, Feuds, and Diamonds: Ebola and the Ravages of History, Paul Farmer (2020). The idea it is built on appears first in his essays, written while the West African epidemic ran. "prevention is being pitted against care in comments about the Ebola epidemic in West Africa", and the consequence is geographic, since "most of West Africa is in a public health desert, which is why Ebola spreads", "as well as a supportive care desert, which is why it kills". The mechanism of death is mundane, because "Ebola kills mostly by causing massive fluid and electrolyte loss", treated by "fluid resuscitation, a fancy term for replacing what has been lost" (London Review of Books 37(3), 5 Feb 2015).

The specific remedy has been tested and does not carry the gap alone. Across treatment units in Liberia and Sierra Leone, "146 (41.3%) cases treated with IVF survived, whereas 31 (44.9%) cases not treated with any IVF survived (P = .583)", with "Matched propensity score analysis found no significant difference in 28-day survival" (Aluisio et al., Clinical Infectious Diseases). What survives is the architectural claim: a test whose only consequence is isolation gets avoided, which is the behaviour behind the late presentations reported from Congo on 2026-10-17.