← Back to all articles

Competition Roundups

Year in Review 2026: How Stripe Young Scientist Winners Turned Big Ideas into Research Questions

What GlioScope, Sheep Strength II and other recent finalists reveal about narrowing a topic, choosing a comparison and making evidence do real work.

16 Nov 202616 min read
Article cover image

Editorial overview

The hardest step in a school research project often comes before the experiment, the code or the survey. It is the moment when an interest has to become a question.

“Brain cancer” is not yet a research question. Neither are “sustainable construction”, “AI”, “flooding” or “noise in schools”. Each is large enough to support a lifetime of work. A student project becomes research only after deciding what will be identified, compared or measured—and what evidence could show that the first idea was incomplete.

The 2026 Stripe Young Scientist & Technology Exhibition made that transition unusually visible. Aoibheann Daly's overall-winning GlioScope project did not present artificial intelligence as a general answer to cancer. Its title defined a disease family, a profiling task and a particular analytical approach. Aoife Fadian and Jessica O'Connor's Best Group project, Sheep Strength II, returned to an earlier idea about wool-reinforced concrete and asked a narrower question about which form of wool could work most effectively and realistically. Rea Vinovrski's award-winning school-noise study moved from an everyday complaint—schools are noisy—to named outcomes involving behaviour, learning and emotional health, with mainstream and SEN students explicitly within scope.

These projects are very different. What connects them is not subject matter but control. The students reduced the distance between the question they asked and the evidence they could realistically collect.

This review therefore does not attempt to rank the most fashionable 2026 topics. It looks at how promising ideas were given boundaries, how the exhibition's recent winners have treated AI and real-world problems, and where an ambitious project can lose focus before the research has properly begun.

Quick Facts

Field2026 exhibition
Edition62nd Young Scientist & Technology Exhibition
Finalists1,060 students from 211 schools
Projects exhibited550: 338 group projects and 212 individual projects
Overall winnerAoibheann Daly — GlioScope: Multi-task Deep Learning and Causal AI for Glioma & Glioblastoma Profiling
Best GroupAoife Fadian and Jessica O'Connor — Sheep Strength II: Using Sheep Wool to Reinforce Concrete
What this review examinesHow broad interests became bounded, testable research questions

For current eligibility, format and official links, see the Stripe Young Scientist & Technology Exhibition competition page. Students planning a future entry can also read How to Prepare for the Stripe Young Scientist & Technology Exhibition.

This review was checked on 14 November 2026.

What the 550 projects actually covered

The official figures give a reliable picture of the broad exhibition. They do not support claims about how many students investigated narrower themes such as AI, cancer, school culture or water quality.

Official categoryProjectsShare of the 550 finalists
Social & Behavioural Sciences14225.8%
Health & Wellbeing11921.6%
Biological & Ecological Sciences10018.2%
Technology9918.0%
Chemical, Physical & Mathematical Sciences9016.4%

The largest category was Social & Behavioural Sciences, not Technology. Health & Wellbeing was close behind. That matters because the exhibition is sometimes imagined as a hall of devices, robots and laboratory demonstrations. In reality, a substantial part of it is concerned with people: how they behave, learn, respond to institutions and experience health.

The winners list adds detail, but it is a selected record. GlioScope appears both as the overall winner and as a partner-award recipient; WaterWatch appears as a partner-award winner and as highly commended. Counting every line on the awards page as a separate research project would therefore double-count some work. The safest conclusion is that 2026 contained a wide range of technical, medical, environmental and behavioural questions—not that any one narrow topic dominated the exhibition.

The winning projects started by choosing what not to study

A broad idea feels ambitious because it contains many possible directions. A research question becomes ambitious in a more useful way: it chooses one direction and follows it far enough to produce evidence.

That act of exclusion is visible in the four main award titles.

GlioScope limited itself to glioma and glioblastoma rather than “all cancer”. It described profiling rather than promising a universal diagnosis or cure. It named multi-task deep learning and causal AI as methods rather than treating AI as a magic label. The public material does not provide enough detail to reconstruct every dataset, target or validation result, and this review will not invent them. But even the official title shows a sequence of decisions: disease, task, method.

Sheep Strength II was equally disciplined in a different field. “Can waste wool be useful?” could lead to insulation, textiles, filtration, packaging or dozens of other projects. The team selected concrete reinforcement. After their 2025 Sheep Strength project became a main-award runner-up, they returned in 2026 to investigate the most effective and market-friendly form in which wool could be used. The second project was not merely “more testing”. It tightened the practical decision the evidence needed to inform.

Joshua Corbett's individual runner-up project, titled “There's Plenty of Room To Break Through at the Bottom”, focused on identifying nanocarriers for intranasal drug delivery in brain-cancer treatment. Ritvik Venkateshwar and Hao Wen Liu's group runner-up simulated Axion Monodromy inflation to investigate whether that specific model could describe the Universe. One works at the scale of drug-delivery particles and the other at the scale of cosmology; both replace a vast field with a defined mechanism that can be examined.

The pattern is worth stating plainly: the strongest question is not necessarily the one with the biggest noun. It is the one that makes the required evidence visible.

GlioScope: define the output before celebrating the technology

Student AI projects often begin with a tool: build a model, create an app, use machine learning. The research question arrives later, if at all. GlioScope's framing reverses that order. The medical problem is named first; the analytical work is described as profiling; the technical approaches follow.

This distinction separates a demonstration from an investigation. A demonstration asks whether something can be made to run. Research asks what it is supposed to identify or predict, what counts as success, and against what reference its performance should be judged.

That does not mean every project requires clinical-scale validation. A school student is constrained by time, access, ethics and data. It does mean the limitation should shape the question. “Can I develop a model that profiles these defined tumour types using these available variables?” is more defensible than “Can AI solve brain cancer?” The narrower sentence is not less serious. It is serious enough to be tested.

GlioScope also illustrates why technical vocabulary must earn its place. “Causal AI” implies a stronger ambition than detecting correlation. Without access to the full report, it would be wrong to claim exactly how that causal component was validated. For future entrants, however, the wording creates a useful standard: if a project uses a term such as causal, predictive, accurate or effective, the method must explain what evidence justifies that term.

Sheep Strength I to II: a research question can survive the exhibition

The most instructive story across the 2025 and 2026 results may be Sheep Strength.

In 2025, Aoife Fadian and Jessica O'Connor were main-award runners-up for using wool to reinforce concrete. They could have treated that recognition as the end of the project. Instead, Sheep Strength II returned to the practical uncertainty left inside the first result: which form of wool would be most effective and market-friendly?

That change matters. “Wool can reinforce concrete” is a proof-of-concept claim. A second-stage question has to deal with comparison and use. The material may be processed in different ways; performance may change; a technically strong option may be unrealistic to source or manufacture. The organiser's account specifically describes the 2026 project as an expansion of the previous work, directed towards the most effective, market-friendly form.

This is what iteration looks like when it is more than repetition. The first study establishes that an idea deserves attention. The next identifies the decision that still cannot be made from the existing evidence.

Students sometimes avoid a follow-up project because they fear judges will see it as unoriginal. Sheep Strength II suggests a better test. If the new project has a new uncertainty, comparator or application decision, continuity can be a strength. “II” is justified when the question has moved.

School noise: turning a familiar complaint into variables

Rea Vinovrski's project began with something students experience every day: noise in school. The award title shows how that experience was converted into a research frame—“An Investigation into the Role of Noise in Irish Schools: Impacts on Behaviour, Learning, and Emotional Health in Mainstream and SEN Students.”

The strength of this framing is also its risk. It names a setting, outcomes and populations, but three outcome families can quickly become too much for one project. Behaviour, learning and emotional health each require careful definition. Mainstream and SEN students should not simply be placed in two boxes without considering sample size, context and the fairness of the comparison.

That tension is useful. A good title can reveal the remaining methodological questions rather than conceal them. What type and level of noise is being compared? At what time? Is the outcome self-reported, observed or measured through performance? What baseline represents a quieter environment? How are differences in classroom activity separated from differences in sound?

The project received both the Department of Education and Youth's Education for Sustainable Development award and first place in the Junior Individual Health & Wellbeing category. Its importance for future entrants is not that every school problem should become a survey. It is that proximity can be an advantage. A student who knows the environment can notice a precise problem; the research design must then prevent familiarity from being mistaken for evidence.

What students have been investigating, 2024–2026

Three years of main awards and selected official results reveal recurring concerns, but they do not form a complete dataset of every entry. The trends below are therefore examples of directions that reached the award lists, not a ranking of the exhibition's most common subjects.

Health moved from emergency response to disease profiling

The 2025 overall winner, ACT {Aid Care Treat}, was developed by Ciara, Saoirse and Laoise Murphy as a medical-assistance tool for emergency response. Its functions included precise geolocation and the transfer of medical information to emergency services. The official project title also described iterative development—a signal that the team was refining a working system around a defined emergency process, not merely assembling health information in an app.

In 2026, GlioScope placed a different part of medicine at the centre: the profiling of glioma and glioblastoma. The winning projects do not show one straight trend towards “more advanced medicine”. They show two ways of narrowing health research. ACT organised information and actions around the first minutes of an emergency; GlioScope organised data analysis around a defined cancer-profiling task.

Other 2026 awards ranged from pancreatic-cancer treatment identification to asthma prediction and emergency communication for deaf and hard-of-hearing people. “Health” was not a method. Each project had to decide whether its output was a candidate treatment, a warning, a communication aid or a profile.

AI was sometimes the object of study and sometimes the instrument

Seán O'Sullivan's 2024 overall winner, VerifyMe, addressed authorship attribution in the post-ChatGPT era. Its official exhibition overview described a pivot away from chasing rapidly changing AI models and towards the characteristics of individual human writers, using stylometry to distinguish authors and detect AI style mimicry. Here, generative AI was part of the problem being investigated.

In 2026, AI played several different roles. WaterWatch used it within an early flood-warning system. GlioScope used deep learning and causal AI for tumour profiling. Other award-listed projects examined gender bias in generated text, images, voice and language, or asked whether AI helped or hindered people with dyslexia. These should not be collapsed into a single “AI trend”.

There is a meaningful difference between researching AI, using AI to investigate another problem and adding AI to a device because the label sounds current. A well-framed project makes that role explicit. If the AI component disappeared, would the research question remain? If so, the project must explain why AI is the appropriate method. If not, it must define precisely which property of AI is being tested.

Environmental ideas became material and system comparisons

The 2025 main awards included PM-DAC, a direct-air carbon-removal system; the first Sheep Strength study; and work on comfrey as a sustainable soil enhancer. The 2026 list contained projects on bioadsorption for fashion runoff, fertiliser amendments for barley, soil organic matter measured with UAV imagery, bioplastics and water purification.

The common thread is not simply sustainability. It is translation. Carbon capture has to become a system whose removal can be measured. Wool has to become a material treatment compared within concrete. Soil improvement has to become a change in selected plant or soil outcomes. An environmental aim supplies a reason for the study; it does not supply the method.

School life became legitimate research terrain

The 2026 awards also show students treating their own social environment as a place where mechanisms can be investigated. Alongside the noise project were studies of adolescent male attitude development, autism labels and peer perception, teacher retention, family vlogging and child-labour law, school uniforms and appearance pressure, social media as escape, and screen time's relationship with memory.

These topics can produce useful research because students have direct access to the questions and language of school life. They can also produce weak causal claims if a small survey is asked to explain an entire social problem. The closer the subject feels, the more important it becomes to distinguish an observation from an explanation.

Where promising research projects lose focus

The topic is mistaken for the question

“Microplastics”, “mental health” or “AI in education” may be an excellent starting interest. None tells the student what to collect. A workable question needs an outcome, a defined context and usually a comparison: which microplastics, affecting what, under which conditions, compared with what?

The project tries to improve everything at once

An app that promises to improve safety, wellbeing, access, achievement and communication has created five evaluation problems. Choosing one primary outcome does not prevent a student discussing wider benefits. It prevents every positive observation from being treated as proof.

There is no comparator or baseline

“The material was strong” means little without another material, an untreated sample or a relevant standard. “Students found the room noisy” is difficult to interpret without a quieter condition, repeated measurements or a defined threshold. A comparator turns description into evidence.

The tool is clearer than the purpose

Projects built around an app, sensor or model can become engineering demonstrations. The question should identify the decision the tool supports and the evidence required to assess it. A flood-warning system, for example, needs a meaningful warning target; the presence of an AI model is not itself the result.

Too many measurements obscure the main result

Collecting more variables can make a project feel scientific while weakening its argument. Every additional measure creates another choice about reliability, sample size and interpretation. The central question should determine the measurements, not the other way around.

Causal language outruns the design

A correlation between screen time and memory does not by itself show that one caused the other. A survey of attitudes does not establish how those attitudes formed. Strong research can still be observational, but its conclusion should stay within what the design can support.

A practical test for a research question

Before writing a proposal, a student should be able to complete five sentences:

  1. I am studying this exact object, population or system.
  2. I will compare these conditions, groups or versions.
  3. My primary outcome is this measurable change or performance.
  4. I can collect credible evidence using this realistic method and access.
  5. My first idea would be weakened if this result appeared.

The fifth sentence is often the most revealing. If no possible result could weaken the preferred conclusion, the project is advocacy or product promotion rather than an investigation.

Consider the difference:

Broad interestMore researchable direction
Wool is sustainableHow do defined forms or treatments of sheep wool compare as reinforcement in a specified concrete mix?
Noise affects studentsHow do measured classroom-noise conditions relate to one clearly defined learning or wellbeing outcome in a specified group?
AI can help medicineCan a defined model perform a specified profiling or prediction task on an appropriate dataset, against a stated baseline?
Flooding is a serious problemCan selected sensor inputs provide a useful warning for a defined flood condition, and how early and reliably?

These are not templates to copy. They demonstrate the work a question must do: name the boundary, the comparison and the evidence.

What 2026 ultimately rewarded

The award list is too varied to support a single formula. A mathematical simulation, a medical AI project, a material experiment and a school-based behavioural study cannot be judged by identical methods. Yet they can be judged by a shared intellectual standard.

The student must know what claim the project is capable of making.

That is why the movement from big idea to research question matters more than a fashionable subject. A broad topic creates excitement, but a bounded question creates the possibility of being wrong. Once that possibility exists, comparison, measurement, revision and genuine discovery can begin.

The 2026 winners did not make their problems small. They made them specific enough to investigate.

Key Takeaways

  • The 2026 exhibition presented 550 finalist projects across five categories; the category totals describe the full field, while award examples should not be treated as a frequency analysis of narrower themes.
  • GlioScope narrowed a vast medical challenge by defining the disease family, profiling task and analytical approach.
  • Sheep Strength II shows that returning to a previous project can be original when the next study addresses a new practical uncertainty.
  • The school-noise project demonstrates how a familiar experience can become research by naming outcomes and populations, while also showing the need to control scope.
  • Across 2024–2026, AI appeared both as an object of investigation and as a tool used inside medical, environmental and social projects.
  • Strong projects define one primary outcome, a comparison or baseline, a realistic evidence source and a result that could challenge the preferred idea.
  • A device, model or app is not automatically a research project; its purpose and evaluation question must be explicit.
  • The most convincing ambition is not a larger topic but a tighter relationship between question, method and claim.

Sources checked

Original CompeteMap editorial content. Brief quotations must clearly credit CompeteMap and link to this article. For complete reuse terms, see Copyright & Reuse.

EXPLORE NEXT

Not sure where to start?

Find the right competition
View all articles →

Loading comments…