The hardest part of a science project isn’t the science. It’s finding a question worth months of your life.
This is the tool I wish I’d had at 16. Answer a few questions and get project ideas built around what you actually care about and what you can actually get your hands on — grounded in what has genuinely won before.
380+
Project ideas
38
Winners to learn from
10
Competitions covered
Free
No account needed
The Lab
Let the quiz build something around you, or bring an idea you already have. Either way you can collect what you like into a set and turn it into one plan. If you’d rather just browse, the idea bank and the library of past winners are further down.
Seven questions, then project ideas built around your answers.
Before we start
It changes what I suggest — and what permission we need before sending you anything.
You can do the whole quiz without giving any personal details — we only ask for an email at the end if you want your summary sent to you.
Idea Bank
Every idea here is a real experiment — a question with something you can measure and something to compare it against. No baking-soda volcanoes, no demonstrations dressed up as research. Across 11 fields, filtered by how much equipment you can actually get hold of.
These are starting points written for this site, not projects anyone has submitted. Take one and make it yours — the best version will be the one you change.
380 ideas · page 1 of 16
How does water turbidity in a local stream change in the hours and days following heavy rainfall?
Measure: Turbidity measured against a printed Secchi/turbidity scale, sampled daily at a fixed point, versus a dry-weather baseline.
Does soil salinity decrease with distance from a winter-gritted road, and at what distance does it return to background?
Measure: Electrical conductivity of soil samples taken at set distances from the kerb, against a control site far from any road.
Does the diversity or coverage of moss species on walls correlate with proximity to heavy traffic?
Measure: Percentage moss coverage in fixed quadrats at varying distances from a main road, versus a rural control.
Does microplastic concentration in river sediment increase downstream of a town or wastewater outfall?
Measure: Particle counts per kilogram of dried sediment after density separation, compared upstream and downstream.
How does surface and air temperature vary between paved, grassed and tree-shaded sites within one town?
Measure: Air and surface temperature at matched times of day across site types, repeated over several weeks.
How does night-sky brightness change with distance from the town centre?
Measure: Sky brightness via a phone light meter or star-count method at fixed distances, controlling for moon phase.
Does carbon-to-nitrogen ratio or turning frequency have a bigger effect on compost decomposition rate?
Measure: Mass loss and internal temperature of replicate compost jars over eight weeks, against an unturned control.
Does irrigation with diluted greywater affect the growth of a fast-growing plant compared with tap water?
Measure: Height, leaf count and dry mass of replicate seedlings across dilutions, versus a tap-water control.
Does the density of dark rubber particulate in puddle water increase with road traffic volume?
Measure: Filtered particle counts under a microscope per litre, correlated with measured traffic counts.
How does UV exposure time affect the tensile strength and fragmentation of common plastic films?
Measure: Force to tear standardised strips after set outdoor exposure times, against strips kept in the dark.
How does habitat type affect the rate of leaf-litter mass loss over a season?
Measure: Mass loss from identical mesh litter bags buried in each habitat, recovered at intervals.
Does rainwater pH vary systematically between sampling sites at different distances from an industrial area?
Measure: pH of rainwater collected in identical clean funnels across sites, over many rain events.
What is the rate of spread of a named invasive species along a river bank or verge over one season?
Measure: Mapped extent using fixed GPS points and quadrat coverage at monthly intervals.
What proportion of material placed in recycling bins is contaminated or non-recyclable, and does signage change it?
Measure: Sorted mass by category before and after introducing clearer signage, across multiple bins and weeks.
How does vegetation cover type affect soil loss under simulated rainfall on a slope?
Measure: Mass of sediment collected from runoff trays with different covers, at constant slope and water volume.
How does CO2 concentration in an occupied room change with ventilation strategy?
Measure: CO2 ppm logged over lessons under different window/door configurations, controlling for occupancy.
Does the invertebrate biotic index of a stream change above and below a known pollution source?
Measure: Kick-sample invertebrate identification scored on a standard biotic index, replicated at each site.
How does partial shading affect the growth of a crop plant compared with full sun?
Measure: Biomass and leaf area of replicate plants under set shade fractions, versus unshaded controls.
How does water acidified with dissolved CO2 affect the mass of calcium carbonate shells over time?
Measure: Dry mass loss of cleaned shells held in water at controlled pH values over several weeks.
How does measured noise level vary across a town by time of day and land use?
Measure: Decibel readings on a calibrated phone app at fixed points and times, mapped and compared.
How does fabric type affect the number of microfibres released in a single wash cycle?
Measure: Filtered fibre counts from wash water per garment mass, comparing polyester, cotton and blends.
How much lower is surface temperature under street-tree canopy than on adjacent open pavement?
Measure: Infrared surface temperatures paired at matched times, across many tree/no-tree pairs.
How does washing duration versus soap type affect the number of bacterial colonies cultured from hands?
Measure: Colony counts on nutrient agar from standardised hand contact plates across conditions, with a no-wash control.
Do extracts of common culinary spices inhibit the growth of a harmless lab bacterium, and which is strongest?
Measure: Zone-of-inhibition diameter on seeded agar plates for each extract, against a solvent-only control.
How does sugar type affect the rate of CO2 production by yeast at constant temperature?
Measure: Gas volume displaced per minute across sugar types, with equal molar concentrations and a no-sugar control.
How does earthworm density affect water infiltration rate and organic matter breakdown in soil columns?
Measure: Time for a fixed water volume to drain through replicate columns, plus mass loss of buried litter.
Do different honey types differ in their inhibition of bacterial growth, and does heating destroy the effect?
Measure: Zone of inhibition per honey type, raw versus heated, against a sugar-syrup control matched for sweetness.
How does surface material affect the amount of biofilm that forms under identical wet conditions?
Measure: Crystal-violet staining absorbance per unit area across materials, replicated.
Do woodlice show a measurable preference for humidity, darkness or substrate when given a choice?
Measure: Proportion of individuals on each side of a choice chamber over fixed intervals, with side swaps to control bias.
How does the number and diversity of seedlings germinating from soil samples vary by land use?
Measure: Seedling counts and morphospecies richness from equal soil volumes grown under identical conditions.
Does flower colour or species have a larger effect on pollinator visitation rate?
Measure: Visits per flower per ten-minute observation window, replicated across times of day and weather.
How do storage temperature, humidity and preservatives compare in delaying visible mould growth?
Measure: Days to first visible colony and percentage coverage over time across replicate slices.
Does lichen community composition on trees vary with distance from a pollution source?
Measure: Lichen type frequency in fixed quadrats on comparable tree species along a transect.
How does buffer composition or incubation temperature affect DNA yield from fruit?
Measure: Visible precipitate mass or absorbance ratio across protocols, replicated on identical fruit.
How does flour type affect the rise rate and acidity of identical sourdough starters?
Measure: Height gain per hour and pH over a feeding cycle, across flour types with matched hydration.
Do soil isolates from different local habitats differ in their ability to inhibit a harmless test bacterium?
Measure: Inhibition zones produced by soil isolates cross-streaked against a BSL-1 test organism.
Does mechanical damage to one plant affect a measurable defence response in a neighbouring plant?
Measure: Leaf toughness or trichome density in neighbours of damaged versus undamaged plants, blind-scored.
How does desiccation duration affect tardigrade revival rate from moss samples?
Measure: Proportion revived after set dry periods, from moss collected at one site.
Does perceived threat measurably reduce inter-individual distance in a shoal?
Measure: Mean nearest-neighbour distance from overhead video frames, before and after a non-harmful visual stimulus.
Do extracts of common aromatic plants reduce insect landing rate compared with a control?
Measure: Landing counts on treated versus untreated surfaces in a fixed observation period, using non-biting insects.
How does bacterial colony count vary by surface type and traffic level across a building?
Measure: Colony counts from standardised swab areas cultured identically, mapped by location and footfall.
How does the size and position of a cut fragment affect regeneration time in planaria?
Measure: Days to visible eyespot regeneration across fragment types, replicated.
How does the spectral colour of supplemental light affect seedling biomass at constant total intensity?
Measure: Dry mass and leaf area of replicate seedlings under filtered light colours, intensity-matched.
Does continuous exposure to sound at a fixed volume affect plant growth compared with silence?
Measure: Height and dry mass of replicate plants in acoustically separated but otherwise identical conditions.
How does growing medium affect water use per gram of biomass produced?
Measure: Water consumed and final dry mass per plant across media, giving grams biomass per litre.
Does inoculation with mycorrhizal fungi affect growth in nutrient-poor soil more than in rich soil?
Measure: Dry mass across a two-factor design of inoculation and soil fertility, replicated.
How does mulch material affect soil moisture retention over time after a single watering?
Measure: Soil moisture readings at fixed depth over days, across mulch types with a bare-soil control.
How does soil salinity affect germination rate across several crop species?
Measure: Percentage germination and seedling length at graded salt concentrations, per species.
Does interplanting an aromatic species reduce herbivory on a target crop?
Measure: Percentage leaf area lost to herbivory in mixed versus monoculture plots, scored blind.
How does pre-soaking duration affect germination speed and uniformity?
Measure: Days to 50% germination across soak treatments, replicated with many seeds per group.
How does coffee-ground amendment rate affect seedling growth and soil pH?
Measure: Biomass and soil pH across amendment rates, versus unamended control.
How do planting density and light intensity interact to affect yield per unit area?
Measure: Total biomass per tray area across a density-by-light factorial design.
Which low-cost covering best maintains temperature around plants overnight in cold conditions?
Measure: Minimum overnight temperature under each covering and subsequent frost damage score.
Do leachates from a suspected allelopathic species inhibit germination of a test species?
Measure: Germination percentage and radicle length in leachate versus water control, dose-graded.
Does elevated CO2 in a sealed chamber measurably increase plant biomass?
Measure: Dry mass in CO2-enriched versus ambient sealed chambers, light and water matched.
How does harvest day affect biomass and vitamin C content across microgreen species?
Measure: Dry mass and titratable vitamin C per tray at staggered harvest days.
How does rootstock affect sugar content in fruit of a grafted plant?
Measure: Refractometer Brix readings across grafted combinations, matched for age and watering.
How does biochar addition affect nutrient leaching from soil columns?
Measure: Nitrate concentration in leachate after fixed watering, across biochar rates.
How does soil lead or zinc concentration vary across urban growing sites?
Measure: Colorimetric or test-kit metal concentration across sites, versus a rural control.
Does excluding pollinators with mesh reduce fruit set in a garden crop?
Measure: Fruit set percentage in netted versus open flowers on the same plants.
How does constant incubation temperature affect germination rate and final percentage?
Measure: Germination curves across temperature treatments with many replicate seeds.
Does irrigation with treated greywater affect crop growth and soil microbial load?
Measure: Biomass plus colony counts from soil, versus tap-water irrigation control.
How do plasticiser ratio and drying temperature affect the tensile strength of starch-based bioplastic?
Measure: Force at break for standardised strips across a two-factor design, replicated.
How do coating type and salt concentration interact to affect the rate of steel corrosion?
Measure: Mass loss and rust area of identical nails over weeks across coatings and salinities.
How do storage temperature and light exposure affect vitamin C loss in fruit juice over time?
Measure: Iodine titration of ascorbic acid at intervals across storage conditions.
How does the useful pH range of anthocyanin extracts differ between plant sources?
Measure: Colour transition points against buffered standards, compared with a commercial indicator.
How does water hardness vary across supply areas in your region?
Measure: EDTA titration or test-strip hardness across many sampled taps, mapped.
How do low-cost catalyst alternatives compare in decomposing hydrogen peroxide?
Measure: Oxygen volume evolved per minute per gram of catalyst, replicated.
How does adsorbent material affect the removal of a dye from solution at fixed contact time?
Measure: Absorbance drop via colorimeter across adsorbents, normalised per gram.
How does ground eggshell particle size affect its acid-neutralising capacity?
Measure: Volume of acid neutralised to a fixed pH per gram, across particle sizes.
How does cross-linking density affect the water absorption capacity of a synthesised hydrogel?
Measure: Grams water absorbed per gram dry polymer across formulations.
How do electrode metal pairing and electrolyte affect the voltage and current of a simple cell?
Measure: Open-circuit voltage and short-circuit current across combinations, replicated.
How does sunscreen SPF rating correspond to measured UV transmission through a thin film?
Measure: UV-sensitive bead colour change or UV meter reading under standardised film thickness.
How does acid type and curing time affect the hardness of casein-based plastic?
Measure: Indentation or scratch resistance across formulations, replicated.
How do solar still design parameters affect freshwater yield per unit of incident energy?
Measure: Millilitres condensed per hour per square metre across still designs.
How effectively do aquatic plants remove dissolved metals from water over time?
Measure: Metal concentration drop by test kit or colorimetry, with plant-free controls.
How does replacing a fraction of cement with a waste powder affect compressive strength?
Measure: Force at failure for standardised cured cubes across replacement fractions.
Can chromatography separate and identify the dyes present in commercial confectionery?
Measure: Rf values compared against known dye standards, replicated.
How does temperature affect the activity of a common enzyme, and where is the denaturation threshold?
Measure: Reaction rate across a fine temperature gradient, with rate curves fitted.
How does solute identity and concentration affect freezing point depression?
Measure: Freezing temperature measured across solutes at matched molality.
Does incorporating a healing agent measurably restore strength after damage?
Measure: Tensile strength before damage, after damage, and after healing time, across formulations.
How does packaging atmosphere affect the rate of oxidation in a fat-containing food?
Measure: Peroxide indicator or sensory rancidity score over time across packaging types.
How does formulation affect the switching temperature of a thermochromic mixture?
Measure: Transition temperature measured on heating and cooling across formulations.
Can density separation plus melting point reliably distinguish common polymer types?
Measure: Classification accuracy against known plastic samples, reported as a confusion matrix.
How does truss geometry affect load-to-mass ratio in identical-material model bridges?
Measure: Failure load divided by structure mass across geometries, replicated.
How do blade number and pitch angle interact to affect power output at fixed wind speed?
Measure: Electrical power output across a factorial design in a controlled airflow.
How does insulating material affect the rate of heat loss from an identical vessel?
Measure: Temperature decay curves fitted to find cooling constant, normalised by material cost.
How does panel tilt angle affect daily energy yield across the year at your latitude?
Measure: Accumulated energy per day across tilt angles, repeated over months.
How does parachute canopy geometry affect descent rate at constant mass and area?
Measure: Terminal descent time over fixed height across canopy shapes, replicated.
How does panel material and thickness affect sound attenuation across frequencies?
Measure: Decibel reduction per frequency band measured through panels in a fixed rig.
How does a magnetic bearing affect the deceleration rate of a spinning rotor?
Measure: Time to stop from fixed initial speed, magnetic versus mechanical bearing.
How do fill fraction and launch pressure interact to determine water-rocket apogee?
Measure: Peak altitude by triangulation across a factorial design, replicated.
How does input power affect the temperature differential and efficiency of a Peltier module?
Measure: Temperature difference and coefficient of performance across input powers.
How does piezoelectric element arrangement affect energy harvested per footstep?
Measure: Charge accumulated in a capacitor per standardised impact across arrangements.
How accurately can a phone plus diffraction grating identify emission lines against known sources?
Measure: Measured wavelengths versus known reference lines, reporting error.
How does a tuned mass damper affect the oscillation amplitude of a model structure?
Measure: Peak amplitude and decay time on a shake table across damper masses.
How does body shape affect drag force at constant frontal area and airspeed?
Measure: Force measured on a balance in a simple wind tunnel across shapes.
How does surface colour and finish affect the internal temperature of identical model enclosures?
Measure: Internal temperature over a day across colours, under matched solar exposure.
How does charge/discharge rate affect capacity fade over repeated cycles?
Measure: Measured capacity every N cycles across C-rates, plotted as fade curves.
How does surface treatment affect the coefficient of friction between two materials?
Measure: Angle of slip on an inclined plane across treatments, replicated.
How does turbine design affect power extracted from a fixed flow rate?
Measure: Electrical output at matched flow across turbine geometries.
How does 3D-print layer orientation and infill affect tensile strength?
Measure: Force at break for standardised printed specimens across a factorial design.
Does a radiative-cooling surface reach a lower temperature than ambient under clear night sky?
Measure: Surface versus ambient temperature overnight, across surface treatments and cloud conditions.
How does shielding material and thickness affect measured signal attenuation?
Measure: Signal strength in dBm inside enclosures of different materials.
How does surface soiling affect photovoltaic output, and how quickly does it accumulate?
Measure: Power output before and after cleaning, tracked over weeks with weather logged.
Does a scale model of building density reproduce measured real-world temperature differences?
Measure: Model surface temperatures across densities, correlated with real-site measurements.
How accurately can a small classifier identify bird species from short audio clips recorded locally?
Measure: Classification accuracy and confusion matrix on a held-out set of your own recordings.
Which factors best predict public transport delays using published open timetable and real-time data?
Measure: Model error against held-out days, with feature importance reported.
How well do official monitoring-station readings predict measurements taken elsewhere in the same town?
Measure: Correlation and error between station data and your own sensor logs across sites.
Do linguistic features alone distinguish reliable from unreliable articles in a labelled public dataset?
Measure: Classifier accuracy on held-out articles, with feature ablation.
How does junction configuration affect simulated average vehicle wait time under real traffic counts?
Measure: Mean wait time across simulated configurations, seeded with your own traffic counts.
How does data type affect the compression ratio and speed of different algorithms?
Measure: Compression ratio and time per algorithm across matched file types.
Does a simple collaborative-filtering recommender reduce content diversity over simulated sessions?
Measure: Diversity metric of recommended items over iterations versus a random baseline.
How does training set size affect digit-recognition accuracy, and where does it plateau?
Measure: Accuracy versus training set size curve on a public dataset.
How does real-world password composition compare with theoretical entropy assumptions?
Measure: Entropy distribution computed over a public breach-derived wordlist, no personal data used.
How do pathfinding algorithms compare in runtime and optimality on real map data?
Measure: Runtime and path length versus optimum across algorithms on OpenStreetMap extracts.
How accurately can an image classifier detect early-stage plant disease from leaf photos?
Measure: Accuracy and false-negative rate on held-out images, by disease stage.
How sensitive are SIR-model outcomes to changes in a single behavioural parameter?
Measure: Peak infection and total cases across parameter sweeps, with sensitivity analysis.
How does simulation runtime and memory scale with qubit count for different circuit types?
Measure: Runtime and memory versus qubit count, fitted to a scaling law.
What proportion of local public-sector pages meet basic automated accessibility checks?
Measure: Automated audit scores across a sampled set of pages, categorised by failure type.
How well can time-of-day and weather data predict household electricity consumption?
Measure: Forecast error on held-out weeks versus a naive baseline.
How has vegetation or built-up area changed locally over a decade in public satellite imagery?
Measure: NDVI or classified land-cover change between matched-season images.
How does input size and memory access pattern affect real sorting performance versus theory?
Measure: Measured runtime versus theoretical complexity across input sizes and distributions.
How does federated averaging affect model accuracy compared with centralised training?
Measure: Accuracy gap between federated and centralised training across client counts.
At what image resolution does optical character recognition accuracy collapse?
Measure: Character error rate versus resolution and noise level.
Do published variant databases agree on the pathogenicity of the same mutations?
Measure: Concordance rate between databases across a sampled variant set.
Does text classification accuracy differ systematically across English dialects?
Measure: Accuracy per dialect group on matched-content test sets.
How accurately can accelerometer data classify a small set of hand gestures?
Measure: Per-class accuracy on held-out sessions from a wearable sensor.
How does phone-viewing angle affect measured neck flexion during normal use?
Measure: Neck angle from side-profile photos at intervals, across seated and standing use.
How does evening screen exposure relate to self-reported sleep latency and quality?
Measure: Sleep diary latency and quality scores across nights, with screen time logged.
How does fluid intake affect reaction time across a normal day?
Measure: Reaction-time test scores at set times under differing hydration, self-experiment only.
How does fabric layer count and material affect particle filtration efficiency?
Measure: Particle counts upstream and downstream of the material in a sealed rig.
How does glazing type affect transmitted UV index across conditions?
Measure: UV meter or UV-bead readings inside versus outside across glazing types.
How do cooling methods compare in reducing temperature of a heated tissue analogue?
Measure: Temperature decay curves in a standardised gelatin analogue across methods.
How does simulated inhalation technique affect aerosol deposition in a model airway?
Measure: Mass deposited in a model airway across techniques, using a placebo aerosol.
How does device placement and gait speed affect step-count accuracy versus manual counting?
Measure: Percentage error versus manually counted steps across placements and speeds.
Does a short intervention improve public understanding of antibiotic resistance?
Measure: Pre/post survey scores with a control group, anonymised and consented.
Do blue-blocking lenses measurably reduce transmitted blue wavelengths, and does that match marketing claims?
Measure: Spectral transmission via phone spectrometer across products versus claims.
How does dressing material affect moisture retention in a wound analogue over time?
Measure: Mass loss from a standardised moist analogue across dressings.
How does measured sodium content in prepared foods compare with label claims?
Measure: Titration or conductivity-derived sodium versus stated label values.
How does caffeine intake affect resting heart rate over the following hours?
Measure: Resting heart rate at intervals on caffeine versus non-caffeine days, self-experiment only.
How does handle geometry affect measured wrist angle and reported comfort during walking?
Measure: Wrist angle from video plus comfort ratings across designs, consented volunteers.
How does communication medium affect the accuracy of symptom description in a simulated scenario?
Measure: Accuracy of transcribed symptom sets across media in a role-play protocol.
What sound levels do people actually choose, and how do they compare to safe-exposure limits?
Measure: Measured output level at users' chosen volume via a coupler, versus published limits.
Does measured CO2 concentration relate to performance on a short attention task?
Measure: Task scores against logged CO2, with ventilation manipulated between sessions.
How do low-cost pulse oximeter readings compare across devices on the same person?
Measure: Agreement between devices via Bland-Altman analysis, self and consented volunteers.
How does toothpaste formulation affect acid-induced mass loss in an enamel analogue?
Measure: Mass loss of eggshell or extracted-tooth analogue after acid exposure across pastes.
What proportion of key local routes are navigable by wheelchair, and where do they fail?
Measure: Audited route segments scored against measurable criteria such as kerb height and gradient.
How do storage temperature and packaging affect bread staling rate?
Measure: Crumb firmness by penetration test over days across a factorial design.
How do dip treatments compare in delaying enzymatic browning at constant conditions?
Measure: Colour change scored from timed photographs against a standard chart.
How does emulsifier type and concentration affect emulsion stability over time?
Measure: Time to visible phase separation and separated volume across formulations.
How do binder type and hydration affect the rise and crumb structure of gluten-free dough?
Measure: Rise height and crumb pore area from cross-section images across formulations.
How does cooking method affect ascorbic acid retention in vegetables?
Measure: Iodine titration of ascorbic acid across methods at matched doneness.
How does brine concentration affect fermentation rate and microbial safety markers in vegetables?
Measure: pH decline curves and visual spoilage rate across brine concentrations.
How does measured spoilage onset compare with printed best-before dates across product types?
Measure: Days to measurable spoilage indicator versus label date, across replicates.
How do plant-based milks compare with dairy in protein content, foaming and curdling behaviour?
Measure: Foam volume and stability, plus curdling pH threshold, across products.
How do packaging method and freezer cycling affect ice crystal formation and moisture loss?
Measure: Mass loss and surface ice coverage over weeks across packaging.
How do grind size and brew time affect extracted solids and estimated caffeine?
Measure: Total dissolved solids by refractometer across brew parameters.
Does a colorimetric proxy correlate with reported Scoville heat across chilli varieties?
Measure: Colorimetric absorbance versus published heat ratings, correlation reported.
How does fermentation temperature affect final acidity and flavour intensity?
Measure: Titratable acidity and pH across temperatures, with blind sensory ranking.
How does measured protein content compare with label claims across products?
Measure: Biuret or Kjeldahl-proxy assay versus label values across brands.
How do container geometry and food arrangement affect temperature uniformity after heating?
Measure: Temperature spread across a grid of points after fixed heating, by arrangement.
How does repeated heating affect oil degradation indicators?
Measure: Viscosity and colour change plus peroxide indicator across heating cycles.
Does a non-destructive measure predict internal ripeness as well as cutting does?
Measure: Non-destructive firmness or acoustic response versus destructive Brix, correlated.
At what reduction level does salt content become detectable in a blind taste test?
Measure: Detection threshold from triangle tests across reduction levels, consented volunteers.
How does packaging mass per calorie vary between shopping strategies?
Measure: Packaging mass per unit food energy across matched shopping baskets.
Does signage wording measurably change recycling bin contamination rates?
Measure: Contamination rate by mass before, during and after each signage condition.
How does background sound type affect performance on a reading-comprehension task?
Measure: Comprehension scores across sound conditions, counterbalanced, consented volunteers.
Does tableware colour affect portion size taken in a naturalistic setting?
Measure: Mass of food served across colour conditions, randomised, with consent.
How does time-on-task affect Stroop-test reaction time and error rate?
Measure: Reaction time and errors across task blocks, within-subject.
Does stated group opinion change individual judgement on an ambiguous task?
Measure: Proportion of judgements shifted toward group answer versus control, fully debriefed.
How does an interruption affect completion time and errors on a sustained task?
Measure: Task time and error rate with and without scheduled interruptions, within-subject.
How do spaced retrieval, rereading and summarising compare for recall after a delay?
Measure: Recall scores at fixed delay across techniques, counterbalanced across material.
How does the number of options affect decision time and satisfaction?
Measure: Decision latency and rated satisfaction across option-set sizes.
Does a told-to-be-performance-enhancing inert drink affect measured endurance?
Measure: Time to exhaustion on a standard task, blinded and consented, with debrief.
How do typeface and spacing affect reading speed and comprehension?
Measure: Words per minute and comprehension across typographic conditions.
Does queue configuration affect perceived and actual waiting time?
Measure: Observed wait times and rated perceived wait across configurations.
How does question wording affect recall accuracy of a witnessed event?
Measure: Accuracy and false-detail rate across question framings, consented and debriefed.
How does self-reported sleep duration relate to next-day mood and reaction time?
Measure: Daily mood scale and reaction time against logged sleep, self-study.
Does exposure to an arbitrary anchor shift numerical estimates?
Measure: Mean estimates across anchor conditions, randomised assignment.
Does communication medium affect willingness to give critical feedback?
Measure: Feedback valence scored blind across media conditions, consented.
Does adding streak tracking affect adherence to a daily habit over four weeks?
Measure: Daily adherence rate across conditions with a matched control group.
At what measured noise level does task performance begin to degrade?
Measure: Performance across controlled decibel levels, within-subject, counterbalanced.
Does changing the pre-selected default affect final choice rates?
Measure: Choice proportions across default conditions in a randomised survey.
How does structural height-to-base ratio affect failure on a shake table?
Measure: Amplitude at failure across model geometries, replicated.
Does a cloud-type observation rule outperform chance at predicting rain within six hours?
Measure: Forecast accuracy versus observed rainfall over many days, against app baseline.
How accurately can shadow-length measurements determine local latitude and solar declination?
Measure: Computed latitude versus known GPS latitude, error reported.
How does observed meteor rate vary with time of night and sky brightness?
Measure: Counts per hour across nights, corrected for limiting magnitude.
How do soil texture and horizon depth vary across land uses in your area?
Measure: Texture by sedimentation and horizon depths from auger samples across sites.
How does river discharge respond to rainfall, and how quickly?
Measure: Cross-section and float-derived velocity to compute discharge, logged against rainfall.
Does indoor radon concentration vary systematically with underlying geology?
Measure: Detector readings across homes grouped by mapped bedrock, with consent.
How does surface albedo affect measured surface temperature across ground cover types?
Measure: Paired albedo and surface temperature readings across cover types.
How accurately can a simple harmonic model predict local tide times versus published tables?
Measure: Predicted versus published high-tide times, error in minutes.
How does gnomon design affect sundial accuracy across the year?
Measure: Deviation from clock time corrected for equation of time, across designs.
How much do local microclimate measurements deviate from the nearest official station?
Measure: Paired temperature, humidity and wind versus station data over weeks.
How does rock type affect mass loss under controlled acidic conditions?
Measure: Mass loss of standardised samples over time by rock type.
How accurately can visual satellite observations reproduce published orbital passes?
Measure: Observed versus predicted pass times and elevations, error reported.
How well does barometric pressure trend predict rainfall within 24 hours?
Measure: Prediction accuracy from pressure trend rules versus observed weather.
How does reflector geometry affect peak temperature achieved by a solar cooker?
Measure: Peak water temperature and time to reach it across geometries, at matched insolation.
How does phase-change material mass affect how long a space stays warm after heating stops?
Measure: Time above a threshold temperature across PCM masses, versus no PCM.
How does substrate type affect power output from a microbial fuel cell?
Measure: Power density over time across substrates, with sterile controls.
How does flywheel mass distribution affect energy stored and retained over time?
Measure: Energy recovered versus input across rotor designs, and decay over time.
How much thermal energy can be recovered from waste warm water with a simple exchanger?
Measure: Temperature rise in the recovery loop and computed energy recovered per litre.
What is the aggregate standby power draw of household devices, and what does it cost annually?
Measure: Measured standby wattage per device, aggregated and annualised.
How do measured lumens per watt compare with packaging claims across bulb types?
Measure: Lux at fixed distance and measured power draw, versus stated values.
How does added thermal mass affect internal temperature variation in a model building?
Measure: Daily temperature range inside identical models with and without thermal mass.
How does feedstock composition affect biogas yield from a small anaerobic digester?
Measure: Gas volume produced per gram of volatile solids across feedstocks.
How much additional daily energy does single-axis tracking deliver over a fixed panel?
Measure: Daily energy yield of tracked versus fixed panels, side by side over weeks.
How do low-cost window treatments compare in reducing heat loss?
Measure: Heat flux or temperature differential across treatments in a controlled box.
How do speed, temperature and terrain affect energy consumption per kilometre?
Measure: Wh per km from logged trip data across conditions, regression modelled.
How do electrode material and electrolyte concentration affect hydrogen production efficiency?
Measure: Gas volume per joule of electrical input across conditions.
How does tethered-wing flight pattern affect tension-derived power extracted from wind?
Measure: Tension and line speed derived power across flight patterns at matched wind speed.
How much usable electrical energy can be recovered from a domestic waste-heat source?
Measure: Energy accumulated per hour across temperature differentials and module counts.
What fraction of household electricity demand could be time-shifted without changing behaviour?
Measure: Proportion of logged load classified as shiftable, modelled against tariff periods.
How does measured urban wind speed distribution compare with the cut-in speed of small turbines?
Measure: Wind speed histogram at roof height versus turbine power curve.
How does ambient temperature affect usable capacity of a rechargeable cell?
Measure: Discharge capacity at controlled temperatures across replicate cells.
How does the density and composition of beach litter change after a storm compared with settled weather?
Measure: Items per 100 m transect sorted into standard OSPAR-style beach-litter categories, counted along the same marked transect in settled weather and again within 48 hours of a storm, surveyed on a falling tide with an adult once conditions have calmed.
Does nitrate concentration in well and spring water vary with distance from fertilised fields?
Measure: Nitrate concentration by colorimetric test strip or spectrophotometer across sites at mapped distances from arable land, with mains water and a deep upland spring as controls, sampled with the landowners' written permission and with individual sites reported anonymously.
How does substrate depth on a model green roof affect the volume and delay of runoff from a fixed rainfall?
Measure: Runoff volume collected and time to first outflow from replicate trays at set substrate depths under a constant simulated rainfall, versus a bare tray of the same area.
How does infiltration rate differ between grass, gravel, block paving and tarmac in the same locality?
Measure: Time for a fixed water volume to drain from a driven ring infiltrometer on soft ground, or from a ring sealed to the surface with plumber's putty on tarmac and block paving, replicated across surface types on comparable soils and with the site owner's permission, against undisturbed grassland as the reference.
Does particulate deposited on leaves decrease with depth into a roadside hedge?
Measure: Mass of particulate washed from a known pooled leaf area and collected on pre-weighed filter papers dried to constant mass, sampled at set depths into the hedge, against leaves of the same species from a hedge far from any road plus a blank filter run each session.
How does the number of insect impacts on a vehicle number plate vary with habitat, season and time of day?
Measure: Insect splats counted on a wiped, gridded plate per 100 km, recorded only on journeys an adult is already making, comparing rural and urban routes matched for speed limit, temperature and time of day, with a wiped zero count logged at the start of every journey.
Does bat activity differ between lit and unlit stretches of the same hedgerow or river corridor?
Measure: Bat passes per hour recorded on a heterodyne detector or a phone fitted with a clip-on ultrasonic microphone (a phone alone cannot pick up bat calls) at paired lit and unlit points on the same corridor, on matched nights for temperature and wind, observed from public paths with an adult present and never approaching or disturbing a roost.
Does the start time and species richness of the dawn chorus differ between noisy and quiet sites?
Measure: First-song time and number of distinguishable species per 20-minute dawn recording, compared between high- and low-decibel sites matched for habitat and measured on the same mornings.
How do orientation, floor level and glazing area affect the number of hours a classroom spends above a comfort threshold?
Measure: Hours above 26 degrees C logged by identical temperature loggers — first run side by side for a day to check they agree — placed in rooms of differing aspect over several weeks with the school's permission, against a shaded north-facing room as the reference.
Does fitting an aerating showerhead reduce litres used per shower, or do people simply shower for longer?
Measure: Litres per shower from measured flow rate multiplied by logged duration, for the same consenting household members before and after the swap, with shower length recorded separately and results reported by anonymous participant code.
Does introducing a written meal plan reduce the mass of avoidable food waste a household throws away?
Measure: Mass of avoidable food waste per person per week, sorted into categories and weighed daily, for four baseline weeks versus four intervention weeks, with the household's informed consent.
How does greenhouse-gas intensity per 1000 kcal vary across the food categories in a real shopping basket?
Measure: kg CO2e per 1000 kcal computed for each item from a published life-cycle dataset, totalled over the researcher's own household receipts used with the household's consent and stripped of names, card and loyalty numbers, and compared with a substituted basket of equal calories and cost.
How have the annual counts of warm nights and frost days changed in your nearest long-running weather station record?
Measure: Counts of nights above and below fixed thresholds per year fitted with a trend line and confidence interval, comparing an urban station with a nearby rural station over the same period.
Does satellite land-surface temperature correlate with vegetation cover across neighbourhoods of one city?
Measure: Mean land-surface temperature against mean NDVI per census area from matched cloud-free summer scenes, tested for correlation and compared with a winter scene as a control.
Do nitrogen dioxide levels at a monitoring station fall during school holidays compared with term time?
Measure: Hourly NO2 from a public monitoring API averaged by hour of day for holiday versus term weekdays, with wind speed, season and weekend days controlled for in the comparison.
Does evening particulate concentration in a village rise as outdoor temperature falls?
Measure: PM2.5 from two identical low-cost optical sensors logged overnight against outdoor temperature, comparing a village with solid-fuel heating to a gas-heated area on the same nights, after running both sensors side by side for one night to check they agree.
Does shaded stream water stay measurably cooler than an open reach during a warm spell?
Measure: Water temperature logged every 15 minutes by identical waterproof loggers in shaded and open reaches of the same stream, paired for depth and flow, over a week of warm weather, with the landowner's permission and loggers secured from the bank.
How much mass do bags labelled compostable or biodegradable lose in a home compost heap over twelve weeks?
Measure: Dry mass loss and remaining surface area of identical bag samples in mesh litter bags recovered at intervals, against conventional polyethylene and untreated paper samples in the same heap.
Does the number of microfibres per litre differ between tap water, bottled water and filtered water?
Measure: Fibre counts on filter membranes under a microscope per litre filtered, with procedural blanks of purified water run alongside every batch to quantify airborne contamination.
How does dawn air temperature vary with elevation along a valley transect on a clear, still night?
Measure: Air temperature at shielded, fixed points down a slope measured within a 30-minute window at dawn, repeated on clear calm nights and compared with overcast windy nights as the control.
Does bud burst in one tree species happen earlier at warmer urban sites than at rural sites nearby?
Measure: Day of year at which a defined bud-burst stage is reached on tagged trees of the same species, plotted against logged site temperature, with rural trees as the comparison group.
How well does a simple elevation-threshold flood model agree with the published flood-risk map for your area?
Measure: Proportion of buildings classified as at risk by a bathtub model on public LiDAR elevation data, compared against the official flood-zone polygons as ground truth, reported as agreement and false-positive rates.
Does a no-idling sign reduce the number and duration of idling engines at a school pick-up point?
Measure: Count of idling vehicles and mean idling duration timed over fixed 20-minute windows for two baseline weeks and two weeks with signage put up with the school's permission, observed from the pavement and recording no identifying details of drivers or vehicles.
Does the wavelength of light shone at one end of a tube change how Daphnia distribute along it?
Measure: Proportion of Daphnia counted in each third of a gradient tube under each filtered wavelength, with an unlit trial and end-swapped repeats as controls.
How does salinity, crossed with incubation temperature, affect the percentage of brine shrimp cysts that hatch within 48 hours?
Measure: Nauplii counted against a known number of cysts in each salinity-by-temperature combination, versus a standard-salinity room-temperature control.
How does the ratio of bran to expanded polystyrene in the diet affect the mass of polystyrene consumed per gram of mealworms per week?
Measure: Dry mass loss of pre-weighed polystyrene blocks per gram of larvae per week across bran ratios, corrected using identical blocks in insect-free containers.
Do snails remove measurably different leaf areas from different plant species offered simultaneously in one arena?
Measure: Leaf area removed per species over a fixed period from photographed discs on graph paper, with identical discs in a snail-free chamber controlling for wilting and curling.
How does the sucrose concentration of a bait affect the number of ants present at fixed times after it is placed?
Measure: Ant counts per bait at five-minute intervals across concentrations, with a water-only bait as control and bait positions rotated between trials.
How does ambient temperature affect the proportion of fruit flies that climb above a marked line within ten seconds?
Measure: Proportion of flies above the mark in replicate climbing tubes at each temperature, against a room-temperature group drawn from the same culture on the same day.
Does the number of soil mesofauna extracted from equal soil volumes differ between land-use types?
Measure: Springtails and mites counted after Berlese-funnel extraction into a moist collector from standard 200 ml samples, replicated per site and released afterwards, with paired sites sampled the same day.
How does a mesh size that excludes soil invertebrates affect mass loss from buried leaf litter over eight weeks?
Measure: Percentage dry mass loss from fine-mesh versus coarse-mesh litterbags buried side by side with equal starting mass, retrieved on the same schedule.
How does soil water content affect the rate of CO2 release by soil microbes in sealed jars?
Measure: CO2 rise per gram of dry soil per hour from a sensor or titration of trapped alkali across moisture levels, with heat-sterilised soil as an abiotic control.
How does the spatial arrangement of oat flakes affect total tube length and number of connections in a Physarum network?
Measure: Total network length and node count traced from time-lapse photographs for each arrangement, compared with the shortest network that connects the same points.
How does substrate type affect the linear rate of mycelial colonisation in identical pasteurised containers?
Measure: Millimetres of mycelial front advance per day from marks on replicate jars per substrate, measured against a standard straw or sawdust substrate as the reference, with uninoculated jars as contamination controls.
Do paired mycelia isolated from different locations form a visible barrage zone more often than pairings from the same isolate?
Measure: Percentage of plate pairings showing a barrage line, plus its width in millimetres, comparing same-isolate pairings with pairings of different isolates of one identified wood-decay species taken from separate logs, scored blind from coded photographs of plates that stay sealed after inoculation.
How does the type of milk affect the percentage mass gain of kefir grains over a 24-hour fermentation?
Measure: Drained grain mass before and after each fermentation across milk types, against a whole-milk treatment run in every round as the reference, with grain lots rotated between treatments so no lot stays on one milk.
How does household detergent concentration affect the frond doubling rate of duckweed over seven days?
Measure: Frond counts per well each day across a dilution series, giving a growth rate and dose-response curve against untreated pond-water controls.
How does dissolved fertiliser concentration affect the growth rate of algae in identical jars of pond water?
Measure: Transmitted-light greyness of each jar photographed every two days against a fixed lamp and printed grey scale across fertiliser doses, calibrated with a dilution series of dense culture, with unfertilised jars of the same pond water as controls.
How do bicarbonate concentration and light intensity compare in their effect on how quickly leaf discs rise in a syringe assay?
Measure: Time for half the discs to float (ET50) in each condition of a bicarbonate-by-light factorial design, with a bicarbonate-free set as control.
Does stomatal density differ between leaves taken from sunlit and shaded parts of the same tree?
Measure: Stomata counted per calibrated microscope field on nail-varnish peels, at least ten fields per leaf, comparing paired sun and shade leaves from the same individual trees.
How does the number of days since imbibition affect amylase activity in germinating barley or wheat seeds?
Measure: Time for the iodine-starch colour to clear in a standard starch solution per gram of seed extract, sampled each day, against a boiled-extract negative control.
How does sucrose concentration in the medium affect the percentage of pollen grains that produce a visible pollen tube?
Measure: Percentage germinated and mean tube length from at least 100 grains per concentration under the microscope, against a sugar-free medium control.
How does the duration of a 90-degree reorientation affect the final curvature angle of seedling roots?
Measure: Root tip angle measured from scanned images 24 hours after each stimulus duration, compared with seedlings kept vertical throughout.
Does individual seed mass predict germination speed and seedling dry mass within a single species and seed lot?
Measure: Seeds weighed individually to 1 mg on a pocket balance and sorted into mass classes, then days to emergence and 14-day seedling dry mass compared across classes grown under identical conditions.
How does the length of a COI barcode fragment affect the accuracy of species assignment against a public reference database?
Measure: Percentage of published sequences correctly assigned to their own species when trimmed to a range of fragment lengths, benchmarked against assignment using the full-length sequence.
How does contact time on copper, stainless steel and plastic affect the survival of a harmless yeast culture?
Measure: Colony-forming units recovered from each coupon after set contact times using identical inoculum volumes, with sterile plastic coupons as the reference surface.
How do sunlight exposure time and bottle material affect the survival of yeast suspended in water?
Measure: Colony-forming units per millilitre after each exposure time in clear PET, glass and foil-wrapped bottles, with the foil-wrapped bottle held at matched temperature as the dark control.
How does cooling rate affect the mean size and size spread of crystals grown from a saturated salt solution?
Measure: Longest dimension of the largest twenty crystals measured from scaled photographs across at least three controlled cooling rates, all from identical starting solutions, with a rapid ice-bath quench as the reference condition.
How does the dye-adsorption capacity of activated charcoal change over repeated use-and-regeneration cycles?
Measure: Milligrams of dye removed per gram of charcoal from absorbance readings after each cycle, compared against fresh unused charcoal and an unregenerated control.
How does the identity of a household acid affect the mass of carbonate scale dissolved per mole of acid supplied?
Measure: Mass loss of matched carbonate chips after a fixed contact time, normalised per mole of acid, across vinegar, citric acid and lemon juice with each acid's concentration first standardised by titration, and with a distilled-water control.
How do pH and storage temperature affect the fading rate of an anthocyanin extract?
Measure: Colour intensity from fixed-lighting photographs converted to RGB channel values at timed intervals across buffered pH levels and two temperatures, with a dark refrigerated reference.
How do salt concentration and humidity affect the rate of patina formation on copper?
Measure: Mass gain of cleaned copper coupons plus green-coverage fraction from photographs over weeks, across salt and humidity treatments with a dry sealed control.
How does film composition affect water vapour transmission through starch and gelatin films?
Measure: Grams of water lost per square metre per day from sealed cups covered with each film at matched thickness, compared with commercial cling film and an open cup.
How does burial in soil affect the mass loss of starch bioplastic compared with commercial compostable and conventional plastics?
Measure: Percentage dry mass loss of matched buried strips retrieved at fixed intervals over twelve weeks, against sterile-sand and unburied controls.
How does mordant choice affect the colourfastness of a plant dye on cotton across wash cycles?
Measure: Colour difference of dyed swatches photographed under fixed lighting after each of ten standardised wash cycles, comparing alum, ferrous sulfate and tannin mordants only (no chromium, tin or copper salts), with an unmordanted swatch as control.
How much acid or base is needed to shift the pH of common beverages by one unit?
Measure: Millilitres of standard acid and base added per 100 mL to move the pH by one unit, measured with a calibrated pH meter, across beverages with distilled water as reference.
How does the measured acetic acid concentration of commercial vinegars compare with their stated strength?
Measure: Titratable acidity per 100 mL from triplicate titrations to a phenolphthalein endpoint, compared with the label claim and a prepared standard of known concentration.
How does the mass of copper electroplated compare with the mass predicted from charge passed at different currents?
Measure: Mass gain of the cathode weighed to 0.001 g against the Faraday prediction from measured current and time, across at least four low-voltage current settings.
At what surfactant concentration does the surface tension of water stop falling, and does it differ between detergents?
Measure: Drop mass from a fixed-bore dropper as a surface tension proxy across a concentration series, with the break point compared between three detergents and pure water.
How do added binders affect the wet tensile strength of handmade recycled paper?
Measure: Force at break of standardised wetted strips loaded with water in a hanging cup, across binder treatments with untreated handmade sheets as control.
How do storage time and humidity affect the carbon dioxide yield of baking powder?
Measure: Volume of gas collected by water displacement per gram of powder, across storage conditions and open packet ages, against a freshly opened packet.
How does desiccant material affect the mass of water adsorbed at a fixed humidity?
Measure: Grams of water gained per gram of dried material over time in sealed containers at a controlled humidity, comparing charcoal, silica gel, rice and salt with an empty dish.
How do metal choice and contact time affect the yield of copper recovered by displacement from solution?
Measure: Mass of dried copper recovered as a percentage of the theoretical yield, across displacing metals and contact times, with solution absorbance before and after as a check.
How do bead diameter and cross-linking time affect the release rate of a dye from calcium alginate beads?
Measure: Absorbance of the surrounding water at timed intervals to give release half-times across bead sizes and setting times, against free dye added directly to water.
How does the presence of a titanium dioxide photocatalyst affect the rate of dye decolourisation under sunlight?
Measure: Absorbance decline of a dye solution over hours of matched sunlight exposure, comparing catalyst-loaded samples with catalyst-free and foil-wrapped dark controls, with the powder handled wet or in a fume hood to avoid inhaling dust.
Does passing hard water through a magnetic conditioner change the mass of scale deposited on a heated surface?
Measure: Dry mass of scale on identical heating elements after a fixed number of matched boil cycles, comparing magnet-treated water with untreated water and a sham non-magnetic fitting.
Do the pseudorandom generators shipped with common programming languages pass standard statistical randomness tests equally well?
Measure: Pass/fail counts and p-value distributions across a fixed battery of randomness tests per generator, compared against a reference generator run on the same test suite.
How closely do first-digit distributions in open public spending records match the Benford distribution?
Measure: Chi-square deviation from the Benford expectation per data.gov spending dataset, compared with simulated datasets drawn from known distributions.
How do path length and clustering in a Wikipedia link subgraph compare with random graphs of the same size and degree distribution?
Measure: Mean shortest-path length and clustering coefficient of the sampled link graph versus matched random-graph and configuration-model baselines.
How does the size of a network's largest connected component fall as nodes are removed at random versus in order of connectivity?
Measure: Largest-component fraction versus proportion of nodes removed, for random removal against degree-targeted removal, on a published infrastructure or transport network.
Which national indicators in Our World in Data explain the most variance in life expectancy across countries?
Measure: Cross-validated R-squared of single- and multi-predictor regressions against a mean-only baseline, with predictors ranked by added variance explained.
Does the share of residents within a short walk of a scheduled public transport service vary with area deprivation?
Measure: Proportion of population inside a fixed walking isochrone of a stop per small area, compared across census deprivation deciles from open government data and against the whole-region average share as a baseline.
How well does a power-law decay describe aftershock rates in published earthquake catalogues, and does the fit depend on main-shock magnitude?
Measure: Fitted decay exponent and goodness-of-fit per aftershock sequence, compared with a constant-rate Poisson null model on the same time windows.
What transit depth and orbital period can a box-least-squares search reliably recover from public stellar light curves?
Measure: Recovery rate of artificially injected transits as a function of depth and period, with the false-alarm rate measured on signal-free NASA archive light curves.
How does a lexicon-based sentiment score compare with a trained classifier on the same public review corpus?
Measure: Accuracy and F1 for both methods on identical held-out reviews, with a majority-class baseline as the floor.
How accurately do function-word frequencies alone identify the author of a public-domain text passage?
Measure: Author classification accuracy on held-out passages versus a chance baseline, plotted against passage length.
Does automatic transcription word error rate differ systematically across accent groups in a public speech corpus?
Measure: Word error rate per labelled accent group on content-matched sentences, compared with the whole-corpus rate, using only released open-licence recordings.
How much does random rather than chronological train/test splitting inflate reported accuracy on time-ordered data?
Measure: Accuracy gap between random-split and chronological-split evaluation of the identical model and dataset, repeated across several public time series.
Which resampling or class-weighting strategy best improves minority-class detection on an imbalanced public dataset?
Measure: Minority-class recall and precision-recall AUC per strategy, all against an untouched-data baseline model on the same held-out split.
How does reducing numerical precision affect a trained image classifier's accuracy, file size and inference time?
Measure: Accuracy, model size and inference latency at each precision level, compared with the full-precision model on an identical test set.
What is the smallest image perturbation that flips an image classifier's prediction, and how does it compare with random noise?
Measure: Perturbation magnitude at first misclassification for optimised versus random noise of equal magnitude, averaged over sampled images from a public test set.
Does choosing which examples to label beat labelling random ones at the same labelling budget?
Measure: Accuracy versus number of labelled examples for uncertainty sampling against random selection, averaged over repeated runs with different seeds.
How much does reported model accuracy vary across repeated random train/test splits as dataset size changes?
Measure: Standard deviation of accuracy across many repeated splits at each subsampled dataset size, against the single-split figure normally reported.
At what tolerance threshold does an agent-based residential model tip into segregation, and does the neighbourhood definition change that threshold?
Measure: Equilibrium segregation index across tolerance values and neighbourhood radii, compared with a control run where agents move at random.
How does neighbour-interaction radius affect the alignment order parameter in a flocking simulation?
Measure: Steady-state order parameter versus interaction radius and noise level, against a control run with the alignment rule switched off.
Does code complexity predict how often a source file is touched by bug-fix commits in public repositories?
Measure: Correlation between per-file cyclomatic complexity and bug-fix commit rate across several open-source repositories, compared against a file-size-and-age-only baseline model to test whether complexity adds anything beyond size.
How do edit-distance and phonetic matching compare at correcting genuine misspellings from a public error corpus?
Measure: Top-1 correction accuracy per method on the identical misspelling list, broken down by error type, with a do-nothing baseline.
How does adding differential-privacy noise trade query accuracy against re-identification risk on a synthetic population?
Measure: Query error and successful re-identification rate versus the privacy parameter, both compared with the unmodified synthetic dataset; no real personal records are used.
How does starting amplitude affect the period of a simple pendulum, and where does it depart from the small-angle prediction?
Measure: Period timed from slow-motion phone video across starting angles from 5° to 60°, compared against both the small-angle formula and the large-angle correction using measured length, with the departure angle reported relative to the repeat-timing uncertainty.
How does ball temperature affect the coefficient of restitution on a fixed surface?
Measure: Rebound-to-drop height ratio from slow-motion video across ball temperatures, versus room-temperature controls on the same surface.
How accurately can double-integrated phone accelerometer data recover the height travelled in a lift?
Measure: Displacement from integrated acceleration versus tape-measured floor-to-floor heights, reporting error growth with journey length.
How does tyre inflation pressure affect the rolling resistance of a bicycle measured by coast-down deceleration?
Measure: Deceleration fitted from phone GPS or accelerometer coast-downs on a flat, traffic-free path, repeated across pressures against a fixed reference-pressure baseline run the same day, with rider mass, surface and wind held constant.
How does spin rate affect the lateral deflection of a ball launched at constant speed?
Measure: Sideways deflection at a fixed distance from slow-motion video, with launch speed and spin rate both measured from the same video for every trial so throws can be compared at matched speed, versus zero-spin launches.
How does the amount and type of soft furnishing affect reverberation time in the same room?
Measure: Reverberation time from decay of a clap impulse recorded on a phone, measured per octave band before and after adding furnishings.
How closely does the Doppler frequency shift of a passing sound source predict its independently measured speed?
Measure: Frequency before and after passing from a phone spectrogram of a constant-tone buzzer carried on the moving source, converted to speed and compared with speed timed over a marked distance, recorded from a safe standoff well back from the roadside.
How does tube diameter affect the acoustic end correction of an open pipe?
Measure: Resonant frequencies swept with a tone generator across tube diameters, with the fitted end correction compared to the standard 0.6r estimate.
How does free length affect the natural frequency of a clamped cantilever, and does it follow the predicted inverse-square relationship?
Measure: Natural frequency from slow-motion video of the vibrating tip across free lengths (or a phone accelerometer whose added tip mass is weighed and included in the prediction), fitted and compared with beam-theory prediction for the same material.
How does temperature affect the viscosity of household liquids measured by a falling-ball method?
Measure: Terminal fall time of a fixed ball through a graduated tube from slow-motion video across temperatures, calibrated against a liquid of known viscosity.
How do head height and tube bore affect siphon flow rate, and how far does it depart from the ideal prediction?
Measure: Volume collected per unit time across head heights and bore diameters, compared against Torricelli-based predicted flow.
How do grain size and moisture content affect the angle of repose of a granular material?
Measure: Repose angle measured from side-on photographs of poured piles across sieved grain sizes and moisture levels, versus dry uniform controls.
How do fin count and spacing affect the cooling rate of an identically heated block?
Measure: Cooling constant fitted from thermocouple temperature decay across fin geometries of equal mass, versus an unfinned block.
How does pinhole diameter affect the resolution of a pinhole camera image at fixed focal length?
Measure: Smallest resolvable line pairs on a printed test chart across pinhole diameters, compared with the diffraction-plus-geometry optimum.
How does bend radius affect light transmitted through a plastic optical fibre?
Measure: Transmitted intensity from a phone lux meter or photodiode with a fixed LED source (not a laser) across bend radii and turn counts, normalised to the straight-fibre baseline.
How does applied load relate to photoelastic fringe order in a transparent plastic specimen?
Measure: Fringe count photographed through crossed polarising films across known applied loads, calibrated then used to predict stress in a notched specimen versus theory.
How do layer height and print speed affect the dimensional accuracy of printed test parts?
Measure: Caliper-measured deviation from nominal dimensions across a layer-height by speed factorial, replicated on multiple prints.
How do target material, surface angle and air temperature affect the accuracy of an ultrasonic distance sensor?
Measure: Sensor-reported distance versus tape-measured distance across target surfaces, incidence angles and air temperatures, with error reported as bias and spread against a flat perpendicular reference target.
How much do low-cost temperature and humidity sensors differ from a reference instrument in accuracy, response time and drift?
Measure: Logged sensor readings versus a calibrated reference thermometer and saturated-salt fixed-point humidity references over weeks, reporting offset, step-response time constant and drift per sensor.
How do cylinder diameter and flow speed determine vortex shedding frequency, and is the Strouhal number constant?
Measure: Shedding frequency counted from video of a dye trace or tell-tale streamer across cylinder diameters and speeds, using a tow-tank pass at controlled speed where a steady flow is not available, tested for constant Strouhal number against the accepted value.
How does breakwater geometry and submergence affect wave height transmitted past it in a model tank?
Measure: Transmitted wave amplitude measured from video against a scale bar across geometries at matched incident wave height, versus an empty-tank control.
Does a numerically integrated quadratic-drag trajectory model predict measured projectile paths better than a drag-free parabola?
Measure: Root-mean-square deviation between predicted and video-tracked trajectory points for both models, across projectiles of differing mass and diameter.
How does simple visual reaction time vary across fixed times of day within the same person?
Measure: Median reaction time over 30 trials of a browser test at five fixed clock times daily for two weeks, each participant compared against their own all-day mean, with informed consent and school ethics approval.
How does measured temperature vary between locations inside a domestic fridge over a normal day?
Measure: Temperature logged every ten minutes at the door, top shelf and crisper for 48 hours, against the 5 degrees Celsius food-safety guideline and a centre-shelf reference probe.
How do container depth and material affect the time a hot food analogue spends between 60 and 5 degrees Celsius?
Measure: Minutes between 60 and 5 degrees Celsius from probe logging in each container geometry, using cooked rice or water as the analogue, compared against a shallow open tray as the reference condition, with the analogue discarded rather than eaten.
How does chopping-board material affect the number of bacteria recovered after an identical cleaning routine?
Measure: Colony counts on nutrient agar from swabbed 25 square centimetre areas after standardised soiling with a BSL-1 culture and identical washing, versus soiled-uncleaned and never-soiled control boards.
How does rinsing bagged salad affect the number of colonies cultured from the leaves?
Measure: Colony-forming units per gram of leaf on nutrient agar after each rinse regime, against unrinsed leaves from the same bag as the paired control, with plates taped shut, incubated at or below 30 degrees Celsius and disposed of without being reopened.
How does contact time before wiping affect the reduction in bacteria recovered from a hard surface?
Measure: Log reduction in colony counts from stainless-steel coupons inoculated with a BSL-1 organism across a series of contact times, against water-wiped and untreated coupons, fitted as a dose-response curve.
How does hand-washing duration affect the percentage of hand surface still carrying a fluorescent tracer lotion?
Measure: Percentage of hand area still fluorescing under UV, scored by image analysis of standardised photographs, across wash durations with an unwashed control, from consented volunteers under school ethics approval.
Does prompt signage change the mass of soap dispensed per person passing a washroom sink?
Measure: Soap mass consumed per counted entry under each signage condition on alternating weeks, against unsigned baseline weeks, under school ethics approval with users notified in advance that observation is taking place and no personal or identifying data recorded.
How does the measured alcohol content of hand sanitisers compare with the percentage stated on the label?
Measure: Alcohol fraction estimated from solution density and evaporative mass loss against a water-ethanol calibration series, compared with the stated label value for each brand, with blind samples of known ethanol concentration run as a method control and all evaporation done under supervision away from ignition sources.
How does backpack load as a fraction of body mass affect forward trunk lean angle during walking?
Measure: Trunk lean angle measured from side-profile video frames at loads of 5, 10 and 15 percent of body mass, against each person's unloaded walk as baseline, with informed consent and school ethics approval.
How does keyboard height affect wrist extension angle and typing error rate?
Measure: Wrist extension angle from side-view photographs plus errors per hundred words typed at each desk height, within-subject against a standard elbow-height control, with informed consent and school ethics approval.
How large is the gap between estimated and device-logged daily screen time?
Measure: Signed difference in minutes between each participant's written estimate and their own phone's logged total for the same day, repeated on three days, anonymised, with informed consent and ethics approval.
Which better predicts next-morning reaction time: total sleep duration or night-to-night variability in bedtime?
Measure: Morning reaction-time scores regressed separately on logged sleep duration and on bedtime standard deviation across four weeks, compared by explained variance, as self-study or with consented volunteers under ethics approval.
How does background noise level affect the volume people choose for the same audio track?
Measure: Output level at the chosen setting measured with a sound-level meter and headphone coupler across controlled background noise levels, against each person's quiet-room choice, with background noise capped at 80 decibels and each exposure limited to a few minutes, under informed consent and school ethics approval.
How do self-reported weekly hours of loud-noise exposure relate to reported temporary hearing symptoms?
Measure: Reported weekly exposure hours against frequency of reported post-exposure ringing from an anonymous questionnaire, compared between year groups, with informed consent and ethics approval.
Do self-reported daily near-work and outdoor hours differ between students who do and do not wear corrective lenses?
Measure: Mean reported near-work and outdoor hours compared between lens-wearing and non-wearing groups from an anonymous survey, reported with confidence intervals, under informed consent and ethics approval.
How does printed text height on medicine packaging affect reading accuracy at a fixed viewing distance?
Measure: Words read correctly at one metre from photographed packaging text of measured x-height, against a large-print control card, with consented volunteers under ethics approval and no medicine handled or taken.
How does footwear type affect step-time symmetry measured from a phone accelerometer?
Measure: Step-time asymmetry index computed from accelerometer traces over a fixed walking course in each footwear type, validated against frame-by-frame video step timing and compared with barefoot walking, with informed consent and ethics approval.
Does regional air-quality data correlate with regional respiratory prescribing rates in published open datasets?
Measure: Correlation between annual mean pollutant concentration and per-capita respiratory prescribing across regions, compared against a model including age structure and deprivation as covariates, using only aggregated public data.
How does appointment-length variability affect total patient waiting time in a simulated clinic schedule?
Measure: Mean and ninetieth-percentile simulated waiting time across scheduling rules and variability levels, benchmarked against a fixed-slot no-overrun baseline over many random runs.
Does asking a general satisfaction question before or after specific detailed questions change the mean general rating?
Measure: Mean rating on the identical general item compared between two randomly assigned split-ballot versions of the same questionnaire, with written informed consent and parental consent for participants under 18.
Does response-scale format, such as 5-point versus 11-point or verbal versus numeric labels, shift the mean rating of the same item?
Measure: Mean and spread of ratings for identical items across randomly assigned scale formats, all anchored to one common reference item, with informed consent and parental consent for under-18s.
Does anonymous versus name-attached response mode change reported rates of mildly undesirable everyday behaviours?
Measure: Reported frequency of behaviours such as missed homework compared between anonymous and identified conditions, with a randomised-response arm as a third comparison, under informed consent, parental consent for minors and school ethics approval.
How do questionnaire length and invitation wording affect completion rate and the item at which people drop out?
Measure: Completion rate and median drop-out item number across randomly assigned length and wording versions sent to the same population, with informed consent on the landing page and parental consent for under-18s.
Does asking people to judge the accuracy of one headline first change their stated willingness to share later headlines?
Measure: Sharing-intent scores for matched true and false headlines compared between randomly assigned prompted and unprompted groups, with informed consent, parental consent for minors and a full debrief naming the false items.
Does attaching a named outlet or a warning label to identical claim text change its rated believability?
Measure: Mean believability rating of identical claims across label conditions, counterbalanced so each claim appears with every label, with informed consent, parental consent for under-18s and a debrief correcting any false claims shown.
How accurately can readers discriminate AI-generated from human-written short passages, and does being told the base rate change accuracy?
Measure: Signal-detection sensitivity d-prime and response bias from hit and false-alarm rates on matched passage sets, compared between groups told and not told the proportion of AI passages, with informed consent and parental consent for participants under 18.
Does gain versus loss framing of numerically identical scenarios change the proportion of people choosing the risky option?
Measure: Proportion choosing the risky option across randomly assigned gain, loss and neutral framings of identical numbers, with informed consent and parental consent for minors.
Does a prior non-refundable token cost change how long people persist at an unsolvable puzzle?
Measure: Time and number of attempts before quitting compared between participants who spent tokens to start and a no-cost control group, with informed consent, parental consent for under-18s, school ethics approval covering the use of deception, and a debrief explaining the puzzle was unsolvable.
How steeply do people discount hypothetical delayed rewards, and does the fitted discount rate differ between age bands?
Measure: Hyperbolic discount rate k fitted to indifference points from a titrating choice task, compared between age bands with bootstrapped confidence intervals, using hypothetical rewards only, with informed consent and parental consent for minors.
Does showing a unit price alongside the pack price improve accuracy in picking the cheapest of two options?
Measure: Percentage of correct cheapest-option identifications and decision time on matched product pairs with and without unit-price labels, within-subject and counterbalanced, with informed consent and parental consent for under-18s.
Does a descriptive-norm message at a collection point change the mean contribution per donor?
Measure: Mean contribution per contributor and contribution rate across randomly rotated message conditions at a school fundraiser, against a no-message baseline period, with organiser permission, school ethics approval, a posted informed-consent notice and parental consent for participating minors.
How does measured door-to-door travel time by bus, bike and walking compare with the times published by journey planners?
Measure: Actual door-to-door minutes per trip against the planner's predicted minutes for the same departure, repeated across peak and off-peak on the same route, with informed consent from any fellow travellers whose trips are logged and parental consent for minors.
How do cycle counts at a fixed point vary with rainfall and temperature relative to total traffic at the same point?
Measure: Cyclists per hour counted at a fixed location against logged rainfall and temperature, expressed as a share of all road users to control for overall demand, using anonymous aggregate tallies with no images or identifying details and school ethics sign-off; any rider interviews use informed consent with parental consent for under-18s.
How much longer are the walking routes people actually take than the shortest path between the same two points?
Measure: Ratio of recorded route length to OpenStreetMap shortest-path length, compared across route characteristics such as lighting, traffic and gradient, using GPS traces donated under written informed consent with parental consent for minors and home endpoints trimmed before analysis.
How much longer is the shortest step-free route than the shortest unrestricted route across a town's pedestrian network?
Measure: Mean detour ratio between step-free and unrestricted shortest paths over several hundred random origin-destination pairs in a routing graph, validated against field-checked ground truth on a random sample of segments, with informed consent and parental consent for under-18s for any accompanying user survey.
Do message length, punctuation and emoji rate differ systematically by recipient relationship in donated message logs?
Measure: Words per message, sentence-final punctuation rate and emoji per 100 words compared across recipient categories within each sender, using logs donated with written informed consent from sender and named recipients, parental consent for under-18s, and anonymisation before analysis.
What proportion of public and commercial signage is multilingual, and does it differ between districts?
Measure: Percentage of surveyed signs carrying more than one language, coded against a fixed rubric by two independent coders with inter-rater agreement reported, compared between districts matched for sign density, photographing signage only and never people, with informed consent for any shopkeeper interviews.
Does the seat a newcomer chooses depend measurably on how full the seating already is?
Measure: Number of empty seats between a newly seated person and the nearest occupant, recorded against current occupancy fraction and compared with a random-seating null model, using anonymous tallies with no photographs or identifying details and prior school ethics sign-off; any follow-up questions use informed consent with parental consent for minors.
How do single-queue and multiple-queue till layouts differ in simulated mean and worst-case waiting time under measured arrival rates?
Measure: Mean and 95th-percentile customer wait time from a discrete-event simulation seeded with your own timed arrival and service rates — anonymous stopwatch timings and headcounts only, no names, photos or personal details, collected with the permission of the canteen or shop manager — compared against a simulation of the layout the site actually uses run at the same measured rates.
How many student clashes and how many exam slots does a graph-colouring algorithm need compared with a real published timetable?
Measure: Number of student clashes and total slots produced by greedy and simulated-annealing colouring of an anonymised subject-enrolment conflict graph (subject combinations and counts only, no student names or identifiers, used with the school's permission), compared with the timetable actually used and with a random-slot-assignment baseline on the same graph.
Does the box-counting fractal dimension of mapped coastline differ systematically between rocky and sedimentary stretches of the same length?
Measure: Box-counting dimension over a decade of box sizes for matched-length coast segments taken from open map data, with a Koch curve of known dimension used to calibrate the method.
How do peak prevalence and final attack rate differ when the same SIR process runs on a real contact network versus randomised versions of it?
Measure: Peak prevalence and final attack rate averaged over hundreds of stochastic runs on a published, already-anonymised empirical contact network, compared with degree-preserving rewired and fully mixed null models built from the same network at identical transmission rates.
How does intercepted message length affect the fraction of letters a hill-climbing frequency attack recovers correctly?
Measure: Proportion of correctly recovered plaintext letters versus ciphertext length across many randomly generated keys and two languages, compared with the recovery rate of random key guessing.
How far do first-digit distributions in published spending, population and river-flow datasets deviate from Benford's law?
Measure: Mean absolute deviation and chi-square statistic of first-digit frequencies for each real open dataset, compared against synthetic samples drawn from an exactly Benford distribution of the same size.
How much shorter is a heuristically optimised collection route than the route currently driven on the same road network?
Measure: Total route distance from nearest-neighbour, 2-opt and simulated-annealing tours computed on real road-network distances, compared with the existing route (from the council's published collection schedule, or one you map yourself) and with the brute-force optimum on small subsets.
Does the degree distribution of a real transport, power or web network fit a power law better than lognormal or exponential alternatives?
Measure: Maximum-likelihood power-law exponent with a bootstrap goodness-of-fit p-value, plus likelihood-ratio tests against lognormal and exponential fits to the same degree data.
How does the probability that a lattice fire spans the grid change with tree density and with lattice geometry?
Measure: Fraction of 1000 simulated ignitions reaching the far edge as a function of density, giving a spanning threshold for triangular and hexagonal lattices compared with the square-lattice case run by the same code and with the published square-lattice site-percolation threshold as a check on the method.
How often do first-past-the-post, Borda count and single transferable vote select different winners from the same preference profiles?
Measure: Proportion of thousands of sampled preference profiles, plus published real ranked-ballot datasets, where each system's winner differs from the Condorcet winner of the same profile.
How well does a Poisson goal-scoring model predict league match outcomes compared with naive baseline predictors?
Measure: Brier score and log-loss on a held-out season for the fitted Poisson model, compared with always predicting a home win and with the league's long-run base-rate frequencies.
How does the false-positive rate of simulated studies rise with the number of tests run and with stopping early to check?
Measure: Proportion of simulated experiments on random noise reaching p below 0.05 as a function of comparison count and peeking frequency, compared with the nominal 5% rate and with a Bonferroni-corrected run.
How does the runtime of trial division, Fermat's method and Pollard's rho scale with the bit length of a semiprime?
Measure: Median runtime over many random semiprimes at each bit length for each algorithm, fitted to a scaling exponent and compared with the algorithm's theoretical complexity.
How quickly do two nearly identical Lorenz-system trajectories diverge, and does that horizon depend on step size or numerical precision?
Measure: Time for the separation between paired trajectories to grow by a factor of e, giving a largest Lyapunov exponent, compared across initial separations, integration step sizes and single versus double precision, and benchmarked against the published Lyapunov exponent for the standard Lorenz parameter values.
Do extreme-value fits to the first and second halves of a long local rainfall record give different 50-year return levels?
Measure: Fitted generalised extreme value parameters and 50-year return level with bootstrap confidence intervals for each half of a published station record, compared against a single stationary fit to the whole series.
The Library
Winning science fair projects from the Google Science Fair, Regeneron ISEF, the Young Scientist & Technology Exhibition, Jugend forscht and the 3M Young Scientist Challenge — each one with the reason it won written out plainly. Filter by competition or by how much equipment you can actually get hold of.
Every one of these was done by someone your age. Read them for the insight — the reason it won — rather than the topic. That’s the part you can borrow.
38 projects · page 1 of 4
The transferable insight
The whole project rests on one piece of ordinary chemistry: plastic and oil are both non-polar, so they stick to each other. Finding a simple physical principle nobody had applied to a famous problem beats inventing something complicated.
How they actually did it
Ran hundreds of repeat trials across ten polymer types, measured before/after concentrations with a home-built spectrometer, and reported the extraction rate per polymer rather than one headline number.
The transferable insight
The science of detecting Ebola already existed — the barrier was the cold chain. Attacking the logistics constraint rather than the detection chemistry is what made it deployable in the places that needed it.
How they actually did it
Adapted an existing lateral-flow assay format, substituted silk fibroin as a stabilising matrix, and tested reagent viability over time at varying temperatures.
The transferable insight
They noticed nodules on pea roots in a garden and asked whether the same bacteria would help an unrelated crop. A curiosity about something ordinary, followed all the way through to a food-security result.
How they actually did it
Tens of thousands of seeds across many controlled trials over months — the statistical power from sheer repetition is what made the result credible.
The transferable insight
Chose a narrow, well-defined mechanistic question inside a huge problem. 'Why does this specific drug stop working?' is answerable by a student; 'cure cancer' is not.
How they actually did it
Cell-culture work in a university lab under mentorship, with the student driving the hypothesis and experimental design.
The transferable insight
Needed no lab at all — a public dataset, self-taught programming, and a genuinely clinical question. One of the strongest routes in for students with no institutional access.
How they actually did it
Wrote the network from scratch rather than using a library, then validated against held-out samples and reported sensitivity, not just accuracy.
The transferable insight
Started from a friend in the Philippines failing school for lack of light at night. A specific person's specific problem produces a much sharper project than a general topic.
How they actually did it
Characterised the voltage available from a realistic hand-to-air temperature gradient, then engineered the circuit around that hard constraint rather than the other way round.
The transferable insight
Replaced an expensive synthetic material with agricultural waste. 'Can I make this out of something free that people throw away?' is a reliably strong project question.
How they actually did it
Compared water retention of her polymer against commercial products and an untreated control, under drought conditions relevant to her region.
The transferable insight
A famous demonstration of persistence: he emailed roughly 200 labs asking for bench space and got one yes. Access is often the real bottleneck, and it is beatable by volume of asking.
How they actually did it
Combined an existing biomarker with an existing sensing material in a new format. Worth noting the approach drew later scientific criticism — a useful reminder that a fair win is a beginning, not a verdict.
The transferable insight
Targeted a property people actually complain about — charging time — rather than the headline metric everyone optimises (capacity). Picking the neglected variable makes a project distinctive.
How they actually did it
Synthesised and characterised the electrode material, then measured charge time and cycle life against a control.
The transferable insight
Triggered by the Flint water crisis in the news. Following a live news story to the underlying technical gap is a repeatable way to find a timely project.
How they actually did it
Paired a chemically sensitive material with cheap consumer electronics and a phone, so the output was a usable tool rather than a bench result.
The transferable insight
Chose to harvest energy sources everyone else ignores because they're small. Combining several weak sources into one useful device is an underused project shape.
How they actually did it
Iterated cheap prototypes and measured output per unit cost — optimising for affordability as an explicit design constraint, not an afterthought.
The transferable insight
A hedgerow outside the house, a real clinical problem (antibiotic resistance), and a method a school lab can run. The ingredients of a winning project are often geographically very close by.
How they actually did it
Standard disc-diffusion assays with proper controls, repeated across plant parts and extraction solvents to find where the active compounds concentrated.
The transferable insight
Pure software, no lab, no materials budget. If you can code, the barrier to a serious project is close to zero — the scarce resource is a well-chosen problem.
How they actually did it
Benchmarked rigorously against established simulators, which is what turned 'I wrote a program' into a measurable scientific result.
The transferable insight
Picked a problem that barely existed two years earlier. Emerging-technology harms are wide open, because there is no established field of adults already working on them.
How they actually did it
Trained on generated datasets and reported detection accuracy against held-out fakes, treating the generator and detector as an adversarial pair.
The transferable insight
Took a thing millions of people do every weekend and asked whether it mattered. Everyday behaviour is under-studied precisely because it seems too obvious to test.
How they actually did it
Held cooking time and temperature constant, varied only the marinade, and sent samples for chemical analysis — a textbook clean single-variable design.
The transferable insight
Combined field measurement with human data to make a policy argument. Projects that end in 'therefore we should do X' land harder than ones that end in a number.
How they actually did it
Recruited asthma sufferers, measured lung function against measured pollutant levels in their own homes, then built a predictive model.
The transferable insight
Rerouted a problem instead of fixing it — if the ear does not work, use the skin. Asking 'what other sense could do this job?' opens up whole project categories.
How they actually did it
Built prototypes and ran listening trials with hearing-impaired participants, measuring perceived musical detail against a control.
The transferable insight
Two years and twelve failed attempts before it worked. The project is a reminder that the winning version is almost never the first version.
How they actually did it
Systematically varied the chemical treatment and curing steps, testing the resulting material for decay and insulating properties.
The transferable insight
Started entirely in silico, which costs nothing but time, and only used lab access once the computer had narrowed half a million options to a handful.
How they actually did it
Virtual screening followed by targeted wet-lab validation — a strategy that makes limited lab access go much further.
The transferable insight
Copied a solution biology had already optimised over millions of years. Biomimicry is a reliable shortcut when your computing or budget is tiny.
How they actually did it
Studied fruit-fly escape responses, implemented an equivalent lightweight algorithm, and benchmarked reaction time against conventional approaches.
The transferable insight
Built it for his own grandfather. Projects aimed at one real person you love tend to be better specified than projects aimed at a market.
How they actually did it
Iterated the sensor for comfort and false-positive rate, then tested it in real overnight use rather than only on a bench.
The transferable insight
Applied a very old, very cheap technology — the slow sand filter — to a modern industrial problem in her own region. Old tech plus new problem is an under-explored combination.
How they actually did it
Ran filter columns over months, measuring degradation rates against untreated controls.
The transferable insight
Found the question inside the routine: dressings get changed on a schedule rather than on evidence. Auditing why a routine exists is a good source of projects.
How they actually did it
Used fractal sensor geometry to maximise sensing area on a small printed surface, then characterised response to moisture.
The transferable insight
Targeted the timing of diagnosis rather than the treatment. Moving detection earlier is often more tractable for a student than curing the disease.
How they actually did it
Worked in a university lab on antibody design and marker binding, with the student defining the diagnostic use case.
The transferable insight
Did not try to beat the best system — tried to match it for one percent of the price. 'Same result, radically cheaper' is a winning frame judges love.
How they actually did it
Trained models on captured driving data and benchmarked detection accuracy against the expensive alternative.
The transferable insight
Used entirely free public data. Genomic databases are open, enormous, and mostly unexplored by students — one of the best no-budget routes into serious biology.
How they actually did it
Trained and cross-validated on curated mutation datasets, reporting accuracy on held-out mutations.
The transferable insight
Simulated first, then proposed a retrofit that costs almost nothing to install. Designing for what can actually be fitted to existing infrastructure beats designing the ideal system.
How they actually did it
Built CFD models of a real cabin geometry and compared contaminant spread with and without the added fin.
The transferable insight
A reminder that the national competitions feed the international ones — winning at home is often the route to the biggest stage.
How they actually did it
Iterated physical prototypes and flight-tested them, measuring stability characteristics against the design predictions.
The transferable insight
Picked a problem defined by danger rather than difficulty. 'Nobody should have to do this' is as good a project motivation as 'nobody can do this'.
How they actually did it
Designed for a hard cost target and real building constraints, then built and tested a working physical prototype.
The transferable insight
Attacked the materials supply chain rather than the performance figure. Constraints nobody else treats as the problem are where the open questions live.
How they actually did it
Built successive prototypes with 3D-printed parts and hand-wound coils, measuring torque and efficiency across designs.
The transferable insight
Astrophysics is one of the few fields where the raw data is fully public and the equipment is someone else's telescope. Access is not the barrier people assume.
How they actually did it
Ran numerical simulations and compared predicted observables with published survey data.
The transferable insight
Found the trade-off nobody had removed: accuracy currently costs the patient radiation dose. Removing a trade-off is more compelling than improving one side of it.
How they actually did it
Trained models on imaging data and validated positional accuracy against the existing radiation-based standard.
The transferable insight
Worked on a threat that does not exist yet. Anticipating a problem before the field does is a way to have a genuinely novel project as a teenager.
How they actually did it
Implemented and benchmarked the system, reasoning explicitly about the threat model rather than just writing code.
The transferable insight
Pure maths needs no equipment, no ethics approval and no consumables — only a genuinely open question and the patience to work on it.
How they actually did it
Proved results formally and verified conjectures computationally.
The transferable insight
Improved the aim rather than the weapon. Precision-of-delivery problems are everywhere in medicine and are unusually tractable with public imaging data.
How they actually did it
Trained on medical imaging datasets and reported tracking accuracy against clinical baselines.
The transferable insight
Her project was already running when the pandemic hit and she pivoted the target. Building a method rather than a single result means you can redirect it when the world changes.
How they actually did it
Used molecular docking software against published protein structures, ranking candidate binders.
The transferable insight
Went around antibiotic resistance instead of through it. When a field is stuck, the opening is often in a completely different mechanism.
How they actually did it
Synthesised nanoparticles and tested antibacterial effect against controls in culture.
The transferable insight
Put diagnosis and treatment in the same cheap device. Combining two steps that normally need two visits is a strong design move for low-resource settings.
How they actually did it
Trained a classifier on eardrum images and prototyped the hardware around it.
What Actually Wins
Having judged, competed in, and lost a few of these, the pattern is remarkably consistent — and it has very little to do with how expensive your equipment is.
Almost every winning project traces back to something specific that bothered someone. Not a subject on a syllabus — a thing in the world that seemed wrong.
"Microplastics" is a topic. "Can a magnetic liquid pull plastic particles out of water?" is a project. The difference is whether you can measure the answer.
One trial is an anecdote. The thing separating a fair project from a demo is replication — and it's the cheapest way to make your result credible.
Some of the best projects were done on a kitchen table. Having no lab forces you to find a simpler mechanism, and simple mechanisms are what judges remember.
Where To Enter
Format, eligibility and paperwork requirements below are a starting orientation, not the official rules — competitions update these most years, so confirm specifics on each one’s own site before you build around them.
Global · Ages 13–18
Ran 2011–2019 and isn't currently open for entry. Kept here because so many of the strongest historic projects in this library came out of it, and its online-first format still holds lessons.
Know before you start
Global · Ages 14–18
The largest pre-college science competition in the world. You qualify by winning an affiliated regional or national fair first.
Know before you start
Ireland · Ages 12–19
Ireland's national competition, running since 1965. Long carried a title sponsor's name (BT, until it stepped down in 2025; Stripe now sponsors it), but the exhibition itself is the constant. Strong tradition of biology and social-science projects with a local hook.
Know before you start
Germany · Ages up to 21
Germany's national youth research competition, founded in 1965. Entries compete in seven fields, from biology to workplace technology.
Know before you start
USA · Ages Grades 5–8
Middle-school focused. Entry is a short video pitching an idea, and finalists are paired with a 3M scientist as a mentor.
Know before you start
USA · Ages High-school seniors
The oldest science competition in the US, running since 1942. Judged on research ability and scientific potential rather than one project alone.
Know before you start
USA · Ages Grades 6–8
Run by the Society for Science, the main US pipeline for middle schoolers — entrants qualify by placing at an affiliated local or regional science fair first. Called Broadcom MASTERS until 2023, so most older advice online still uses that name.
Know before you start
Europe · Ages 14–20
The EU-wide competition students reach after winning their own country's national fair — EUCYS for short. Ireland's Young Scientist Exhibition and Germany's Jugend forscht both feed into it.
Know before you start
Canada · Ages Grades 7–12
Canada's national fair, reached through regional fairs across the provinces. Run by Youth Science Canada.
Know before you start
UK · Ages 11–19
The UK's largest youth STEM event. The judged project strand within it is The Big Bang Competition (formerly the National Science + Engineering Competition), run by EngineeringUK.
Know before you start
The Guides
Having a good idea is the easy half. These are long, honest guides to the parts that actually decide how a project lands — written from having got most of them wrong first.
Common Questions
Start from something that annoys you rather than from a subject you like. Almost every award-winning project traces back to a specific irritation — a stream that smells, a grandparent who can't open jars, a friend with no light to study by. Then narrow it until you have a question you can measure the answer to. The quiz on this page does exactly that: it asks what bothers you, what equipment you can reach, and how long you have, then proposes four projects built around those answers.
In practice: a sharp question rather than a broad topic, a fair comparison against a proper control, and enough repeat trials that the result isn't luck. Judges are far more impressed by a simple experiment done rigorously than an ambitious one done once. Expensive equipment matters much less than most students assume — several projects in the library below were done at a kitchen table.
Yes, and some of the strongest projects have none. Software, data analysis, modelling and simulation projects need only a computer, and public datasets in genomics, astronomy and medical imaging are enormous and largely unexplored by students. Filter the library below by 'Computer only' or 'Kitchen-table level' to see real examples that won at the highest level with no laboratory access.
Plan for around three months if you can. The single most common failure is leaving no time for repeat trials, so build replication into the schedule from the start rather than treating it as optional. If you only have a few weeks, deliberately narrow the question — a small question answered properly beats a big one answered once.
Email a lot of them. Be specific about what you want to do, why their work in particular is relevant, and exactly what you're asking for — bench time, a reagent, twenty minutes of advice. Expect most to not reply; that is normal and not a judgement of you. Jack Andraka's ISEF-winning project famously came after roughly two hundred rejections and one yes.
It depends where you live. Regeneron ISEF is the largest international fair but you qualify through an affiliated regional or national fair first. The Young Scientist & Technology Exhibition (sponsored by BT for 25 years, now by Stripe) is Ireland's national competition, Jugend forscht is Germany's, and the 3M Young Scientist Challenge is aimed at US middle schoolers. Start with whichever your school already enters, because the qualification path matters more than the prestige.
Yes, entirely, with no account and no sign-up. It was built by Fionn Ferreira, who won the Google Science Fair Grand Prize in 2019, as the tool he wishes had existed when he was starting out.
Beyond The Fair
The project matters less than what you do next with it. Competitions, fellowships, grants and programmes exist specifically for young people doing this kind of work — and most students never find out about them in time.
A running list of programmes, competitions and funding open to young changemakers. Worth checking before you finish your project, not after — a lot of them want to see work in progress.
Browse opportunitiesOne last thing
I entered my first science fair at 12 and lost. I kept entering. The project that eventually won the Google Science Fair started because I pulled a plastic-covered rock out of the water near my house and got annoyed about it.
None of it required a lab, a mentor, or being the cleverest person in the room. It required picking something I actually cared about and then being stubborn about measuring it properly. Everything on this page is an attempt to shortcut you to that starting point.
Use the ideas as a starting point, not an answer. The best version of your project is the one you change your mind about halfway through.
— Fionn
Google Science Fair Global Grand Prize, 2019
Science Fair Idea Lab · fionnferreira.com