The rationale behind the initiative rests on a well-documented structural problem in scholarly publishing. Positive results, meaning those that support a research hypothesis, are more likely to be written up, submitted and accepted than null or negative results. The authors of the editorial cite a broad body of evidence showing that this bias operates at every stage of the scientific process, from study design through data collection and analysis to interpretation, and that it can ultimately produce false positive findings. Because career progression and funding decisions reward publication counts and citation metrics, researchers have strong incentives to chase publishable, positive outcomes rather than to document the dead ends they encounter along the way.

The consequences of this bias are more than cosmetic. When errors and null results go unpublished, other scientists may repeat the same failed experiments, wasting time, effort and public resources. A widely used model may carry defects that are common knowledge inside the institute that developed it but invisible to external users of its output. A Springer Nature survey of roughly 11,000 researchers, cited in the editorial, found that only 15 percent of respondents were aware of journals that actively encourage the submission of null results, while 69 percent believed such results would not be accepted. Researchers, in other words, often do not know where to send the lessons they have learned, and so those lessons are shared only informally, in what the editorial describes as a body of knowledge that has only been shared at water coolers in academic hallways so far.

The new LESSONS format comes in two flavours. LESSONS Reports are short, peer-reviewed journal articles that follow the public peer review model used across EGU journals, in which reviews and author responses are published openly. They are deliberately constrained to roughly 1,000 to 2,000 words in the main text, with a commensurate number of figures, tables and references, a length intended to lower the barrier to submission in the same way as existing short formats such as Letters or Geoscience Communication Insights. LESSONS Posts, by contrast, are brief, non-peer-reviewed preprints of typically fewer than 500 words and a single figure, hosted on EGUsphere, the EGU’s interactive community platform. Posts are intended for early-stage reporting, to stimulate discussion or to attract co-authors for a fuller Report, and are checked only by an EGUsphere moderator for basic scientific quality and standards of civil discourse.

Both formats are collected in a new, ongoing interjournal compilation, which the editorial presents as a key design decision. Earlier attempts to publish failures have often foundered. The EGU journal Hydrology and Earth System Sciences once ran a manuscript category called Black Swans and Scientific Falsifications, focused on unforeseen, high-impact events, but it was discontinued after several years without submissions. The editorial attributes that failure to a narrow scope, high expectations, availability in only a single journal and limited advertisement. Dedicated journals elsewhere have struggled too, with the Journal of Negative Results in BioMedicine discontinued and the Journal of Trial and Error operating outside the mainstream. By embedding LESSONS within established journals rather than launching a standalone venue, the EGU ensures that every contribution meets the same scientific standards as regular articles while gaining the visibility of a dedicated compilation.

The scope of the new article type is broad, and the editorial is careful to stress that the words in the LESSONS acronym are illustrative examples rather than exclusive categories. Errors, for instance, can include the misuse of a widely adopted method, such as a study showing that a commonly used 31-day running window introduces a bias into estimates of expected extreme frequency, or hardware that does not behave as its documentation promises, such as a comparison of 16 commercial solar regulators whose power consumption diverged substantially from manufacturers’ specifications. Limitations and shortcomings cover methods applied beyond their valid boundaries, poor calibration, and software bugs in community codes, provided the bug constitutes a genuine learning moment for many scientists.

Surprises form another rich vein. The editorial recounts how ecologists studying salt-marsh seedling establishment noticed ragworms pulling seedlings into the sediment, a chance observation that ultimately revealed the worms play a prominent role in seedling establishment and even garden their own food. In a similar spirit, researchers testing relationships between river discharge and plastic transport found their only strong correlation was with floating organic material, chiefly water hyacinths, which turned out to entangle plastic pollution so effectively that they can now be used strategically to extract plastics from rivers. Null results from carefully controlled studies also qualify, as do demonstrations of non-reproducibility after substantial effort, and so-called origin myths, in which authors trace how a widely used model, metric or method acquired a misunderstood or oversimplified history.

Quality control remains central. LESSONS Reports are held to the same criteria of scientific significance, quality and presentation as any other journal article: the theory must be sound, the methodology appropriate, the data properly analysed and the conclusions logically drawn. Crucially, the editorial insists that the quality of a study is separate from its results, so a null outcome does not in itself signal weak science. The insight must be novel, meaning not formally described in the literature before, and must offer a takeaway that can caution or benefit other researchers beyond the author’s own group. Entirely predictable errors or simple careless mistakes do not qualify, unless they are believed to be common in the community and a review would be useful. Reviewers are given specific guidance, including checking whether the failure could reasonably have been anticipated and whether others are likely to repeat the same approach.

The editorial also situates LESSONS within a wider movement to reform research assessment. The EGU has endorsed the declaration of the Coalition for Advancing Research Assessment, which calls for moving beyond traditional publication and citation metrics, and the new article type aligns with the union’s existing commitment to transparency, exemplified by its public peer review model and mandatory documentation of manuscript histories. Unlike registered reports, which guarantee publication of negative results by reviewing study protocols before data are collected, LESSONS focuses specifically on the reporting of errors and lessons after the fact, a niche that pre-registration does not cover. The authors also argue that publishing failures can give long-overdue recognition to field researchers, who are more likely to generate null results, and to model developers who spend time finding and fixing bugs rather than chasing new short-term features.

Whether the initiative succeeds will depend on the community embracing it. The editorial points to sustained interest in dedicated EGU General Assembly sessions on failures, including the interdisciplinary BUGS session on Blunders, Unexpected Glitches and Surprises, which attracted 21 abstracts in 2025 and 25 in 2026, as evidence of appetite. The authors close with a direct appeal: submit a LESSONS article, take part in the public discussion on EGUsphere, and encourage local journals to join the compilation. If the format takes hold, the scientific record could become not only more honest but more useful, capturing the full arc of trial and error on which scientific progress actually depends, rather than only its triumphant final chapters.

Subject of Research: A new EGU article type for publishing non-positive results, errors and lessons learned in geoscience research

Article Title: Editorial: Introducing a new article type: Limitations, Errors, Surprises, Shortcomings and Opportunities for New Science (LESSONS)

Article References: Proske, U., Hillier, J., Gaillard, S., Blume, T., Queiroz Alves, E., Buiter, S., Carslaw, K. S., von Elverfeldt, K., van Emmerik, T. H. M., Ervens, B., Hut, R., Illingworth, S., Klotz, D., & Pyschik, J. (2026). Editorial: Introducing a new article type: Limitations, Errors, Surprises, Shortcomings and Opportunities for New Science (LESSONS). Geoscience Communication, 9(3), 371-383. https://doi.org/10.5194/gc-9-371-2026

Image Credits: AI Generated

DOI: 10.5194/gc-9-371-2026

Keywords: LESSONS, EGU, publication bias, null results, open science, peer review, geoscience, scientific publishing, reproducibility, research integrity, EGUsphere, negative results

A new study published in Data Mining and Knowledge Discovery by Saurabh Sharma and Joydeep Chandra of the Indian Institute of Technology Patna, together with Souvik Chowdhury of Jadavpur University, proposes an elegant way out of this trap. Their approach, called Memory Augmented Self Distillation, teaches a graph neural network to learn from its own history. Rather than relying on a separate, fully trained teacher model to guide a smaller student, the framework builds a memory of the network’s past states and draws on that stored knowledge to regularize and refine its current training. The result is a measurable improvement of 2.5 to 6 percent in accuracy across a range of benchmark datasets compared with existing graph neural network training and self-distillation methods.

To understand why this matters, it helps to look at the technique the new method builds upon: knowledge distillation. First popularized by Geoffrey Hinton and colleagues in 2015, knowledge distillation is a compression and regularization strategy in which a large, well-trained teacher network transfers its knowledge to a smaller student network. The student learns not only from the ground-truth labels but also from the teacher’s softer, richer output distributions, which encode subtle information about how confident the teacher is and which classes resemble one another. In domains like computer vision and natural language processing, distillation has become a standard tool for building compact, robust models.

Applying distillation to graph neural networks, however, has proven surprisingly difficult. The core obstacle is a phenomenon known as oversmoothing. Graph neural networks learn by passing messages along edges, allowing each node to aggregate information from its neighbors. When the network is deep, this repeated aggregation causes the representations of all nodes to converge toward indistinguishable similarity, effectively washing out the distinctive features that make classification possible. A teacher graph neural network that has been trained to convergence may therefore produce representations that are smooth but information-poor, offering the student little of value to learn from. Conventional teacher-student distillation, so effective elsewhere, struggles to deliver meaningful guidance in the graph setting.

Self-distillation, in which a network serves as its own teacher, sidesteps the need for a separate teacher but runs into a different problem: the information bottleneck. Because the student and the teacher share the same architecture and training data, the knowledge available to transfer is limited by what the model already contains. Without an external source of diverse information, self-distillation can become an echo chamber, reinforcing the model’s existing biases rather than correcting them. Previous efforts to make graph distillation work have explored multi-teacher setups, adversarial distillation, and structure-aware multilayer perceptrons, but each carries its own computational or methodological trade-offs.

The framework introduced by Sharma, Chowdhury, and Chandra takes a fundamentally different route. Instead of a single teacher or a fixed set of teachers, the method constructs a memory bank that captures diverse snapshots of the learning process as it unfolds. These memory entries act as multiple, heterogeneous knowledge sources drawn from the student model’s own trajectory through training. Because each snapshot reflects a different stage of learning, the memory collectively encodes a richer and more varied body of knowledge than any single model state could offer, directly addressing the information bottleneck that plagues conventional self-distillation.

The crucial question then becomes which of these stored sources the network should listen to at any given moment. Listening to a poorly trained early snapshot could mislead the model, while relying exclusively on the most recent state would recreate the echo chamber problem. The researchers solve this with a competency-based knowledge source selection mechanism. This mechanism dynamically evaluates how competent each memory source is relative to the current learning objective and selects the most pertinent one for distillation at each step. In effect, the network continuously asks which of its past selves has the most useful lesson to teach, and adapts its supervision accordingly. This adaptive selection transforms the memory from a static archive into an active, evolving curriculum.

The technical payoff of this design is twofold. First, the distillation signal from competent memory sources acts as a powerful regularizer, discouraging the network from drifting into the overconfident, overfit solutions that plague graph learning on limited labeled data. Second, because the memory sources are diverse, the student is exposed to a broader distribution of knowledge than it could generate on its own, improving the quality of its learned representations. The authors demonstrate these gains across multiple benchmark datasets, including widely used citation networks such as Cora, Citeseer, and Pubmed, as well as graph kernel benchmarks and an elliptic Bitcoin transaction dataset used for anti-money laundering research, where only a subset of classes was analyzed for the experiments.

The evaluation methodology reflects careful statistical practice, with paired t-tests used to establish the significance of the improvements over baseline methods. The comparisons span the landscape of graph neural network architectures, including graph convolutional networks, graph attention networks, and jumping knowledge networks, alongside recent distillation frameworks designed specifically for graphs. The consistency of the accuracy gains across datasets and architectures suggests that the benefit stems from the underlying principle of memory-augmented self-supervision rather than from tuning to any particular benchmark. The final student models also exhibited better generalization, retaining their performance advantages when evaluated beyond the training distribution.

Beyond the immediate results, the study points toward a broader shift in how researchers think about the training of graph-based models. The idea that a model’s own training history is a resource worth preserving and mining is a departure from the standard paradigm in which intermediate states are discarded the moment a new set of weights is computed. It resonates with a simple intuition: lessons from the past, properly curated, can guide better decisions in the present. For graph neural networks, whose vulnerability to overfitting and oversmoothing has limited the depth and reliability of the models practitioners can deploy, that intuition now has concrete, quantified support. As graph learning continues to expand into finance, chemistry, recommendation systems, and network security, techniques like Memory Augmented Self Distillation could become a standard component of the training pipeline, helping models not only to learn from data but to learn from themselves.

Subject of Research: A memory-augmented self-distillation framework for improving the generalization of graph neural networks

Article Title: Taking lessons from history: Memory Augmented Self Distillation for graph neural networks

Article References: Taking lessons from history: Memory Augmented Self Distillation for graph neural networks. (n.d.). https://doi.org/10.1007/s10618-026-01258-z

Image Credits: AI Generated

DOI: 10.1007/s10618-026-01258-z

Keywords: graph neural networks, knowledge distillation, self-distillation, overfitting, oversmoothing, memory augmentation, node classification, regularization, deep learning, machine learning, Taking, lessons

The study, conducted within the framework of the larger Swedish randomized controlled CGA Swed trial, set out to explore how frail older people managed everyday life in quarantine during the first year of the pandemic. Frailty, in the clinical sense, refers to a state of reduced bodily function, increased vulnerability, and diminished ability to handle stressors, often assessed through criteria such as general weakness, reduced endurance, weight loss, low physical activity, and slow walking speed. Because frailty markedly increases after the age of 80 and is associated with reduced quality of life, impaired cognition, and heightened risk of depression and pain, the researchers reasoned that this group would experience pandemic restrictions differently from healthier older adults. Yet, until now, in-depth knowledge about the consequences for frail older people specifically remained scarce, leaving a gap in both gerontology and occupational therapy research.

Twenty participants, aged 78 to 100 years and including 11 women and 9 men, were recruited through strategic sampling designed to capture heterogeneity in age, illness, functional ability, and dependence in activities of daily living. Almost all depended on home-help services, two lived in nursing homes, and seven rated their health as good while the rest described it as fair or poor. Because meeting in person posed an unacceptable infection risk, the researchers conducted semi-structured telephone interviews between October 2020 and February 2021, lasting from 11 to 33 minutes with a median of 23 minutes. Notably, recruitment and data collection never occurred digitally, an advantage the researchers highlight given that digital exclusion is itself a barrier for this population. The interviews were transcribed verbatim and analyzed using the qualitative content analysis method described by Graneheim and Lundman, in which meaning units are coded, grouped into categories, and distilled into themes, ultimately yielding one overarching theme and four subthemes.

That overarching theme, everyday life in double exposure, captures the central insight of the study: even before the pandemic, limitations already existed across multiple domains of these participants’ lives as consequences of ageing and illness. Dependence on elder care services could negatively affect control over daily routines, and most participants had already adapted their leisure activities to a reduced capacity. The loss of friends to old age had narrowed social interactions, and reliance on relatives for support had heightened dependence. The restrictions then compounded every one of these pre-existing limitations. Rather than confronting the pandemic from a baseline of healthy autonomy, these older people faced a crisis that amplified vulnerabilities they were already struggling to manage, and their diminished capacity to generate new coping strategies made the disruption especially destabilizing.

The first subtheme, powerlessness under the restrictions, describes the tension between adhering to public health guidance and struggling with its suffocating impact. Participants spoke of the pandemic period as a lost year, a time of living unnaturally under a new set of rules about how to exist. The constraints heightened their awareness of life’s finite nature and intensified the desire to engage in meaningful occupations, while simultaneously undermining their sense of autonomy and occupational identity. For those in care homes, limitations were particularly severe, with periods when residents could not move through common areas or meet relatives, and when fellow residents died from the virus. One participant described a common room where only three grieving residents remained after many others had died at the same time. Relatives’ opinions also shaped what activities participants felt permitted to pursue, an external control that reinforced feelings of limited agency and left some waiting for something to change without knowing how to influence their situation.

To manage this powerlessness, participants adopted a strategy the researchers termed acceptance and trust in the restrictions. Accepting the necessity and effectiveness of the measures became a way to cope with double exposure, allowing participants to focus on external factors such as vaccination prospects or changing seasons. Strict adherence, including avoiding all social contact, provided a sense of security for some, while others balanced precautions such as masks, sanitizers, and physical distance with continued essential activities. Healthcare was still sought when needed, and routine visits continued. Yet the study uncovered a novel and troubling finding: inconsistent adherence to safety measures among elder care staff, such as variable use of protective equipment, undermined participants’ sense of security and even led some to cancel support services they clearly needed, despite having extensive care needs that limited that option. Over time, pandemic behaviors like distancing and abstaining from handshakes became routine, but adherence occasionally relaxed, with some participants prioritizing physical contact with close relatives over fear of infection, weighing emotional needs against physical risk.

The third subtheme, limitations in everyday life beyond control, documents how the pandemic profoundly restricted activities and deepened boredom. The inability to meet relatives, socialize with friends, or leave the home for everyday societal activities such as shopping, using public transport, or simply sitting on a bench was experienced as particularly restrictive. Shifting social and religious activities to digital platforms was often impossible due to limited technological skills or resources, meaning that lost participation also meant lost routines, health benefits, and intellectual engagement. Dependence on family members introduced further constraints, as feelings of being a burden inhibited participants from expressing their own needs. Practical frictions multiplied; one participant described the difficulty of coordinating a hairdresser visit with home care schedules that had themselves become unpredictable because of pandemic staffing uncertainties. One participant, asked what she did when spontaneous meetings were no longer possible, responded through tears that she simply lay down in bed and fell asleep, calling the situation awful.

Against these losses, the final subtheme, striving for meaningfulness in everyday life, reveals remarkable resilience. Participants adapted leisure activities to fit within restrictions, meeting friends outdoors when weather permitted, maintaining phone contact, and turning to solitary hobbies such as reading, puzzles, radio, crosswords, and television. Some expanded their use of smartphones and tablets to access social media or games, while others relied on relatives for basic digital tasks, but for many, technology remained inaccessible due to lack of access or competence. Crucially, even participants with digital tools did not perceive online interaction as a true replacement for in-person contact, consistent with prior research on occupational disruption. In care settings, the social interaction provided by care staff grew increasingly important, and some participants supported peers facing greater challenges, demonstrating solidarity, such as one resident who gave fruit to and talked with two women who died the very same night.

The authors conclude that frail older people may have limited capability to adapt to new circumstances, make active decisions to influence their situation, and proactively manage their health, which underscores the central role of occupational therapy in mitigating the impact of disrupted everyday life. Occupational therapists, the study argues, are uniquely positioned to identify how frailty constrains activity performance and participation, to co-create modified routines, to enable safe engagement in daily activities, to strengthen social participation through both digital and non-digital means, and to work with families and care staff to stabilize expectations around daily roles. Addressing digital exclusion, ensuring stable and predictable care environments, and strengthening continuity in primary care through designated contact persons emerge as concrete priorities. Perhaps most significantly, the mechanisms identified here, including reduced activity opportunities, disrupted roles, and weakened belonging, operate not only during pandemics but also in ordinary circumstances where illness, mobility limitations, or restricted social networks chronically isolate frail older people. The study’s findings thus serve as both a record of a lost year and a blueprint for sustaining leisure, connection, and wellbeing in future disruptive events.

Subject of Research: Managing Everyday Life in Double Exposure: Frail Older People’s Experiences During a Pandemic

Article Title: Managing Everyday Life in Double Exposure: Frail Older People’s Experiences During a Pandemic

Article References: Sjöberg, K., Dahlin-Ivanoff, S., Wilhelmson, K., & Andersson Hammar, I. (2026). Managing Everyday Life in Double Exposure: Frail Older People’s Experiences During a Pandemic. Scandinavian Journal of Occupational Therapy, 33(1), Article 1. https://doi.org/10.1007/s44474-026-00001-7

Image Credits: AI Generated

DOI: 10.1007/s44474-026-00001-7

Keywords: Managing, Everyday, Life, Double, Exposure, Frail, Older, People, Experiences, During, Pandemic, scientific research