Safety, ethics and the paperwork that disqualifies people.

The One Rule

Approval comes before
experimentation. Always.

If you remember one sentence from this page, make it this one: at any fair affiliated with the International Science and Engineering Fair, certain kinds of research must be reviewed and approved in writing, before you begin experimentation, and there is no such thing as retroactive approval. Nobody can sign a form later to cover work you already did. If they could, the review would mean nothing, so the rule is absolute and it is applied to projects that are otherwise excellent.

The categories that trigger prior review are consistent year to year even as the forms change: human participants, vertebrate animals, potentially hazardous biological agents, human and animal tissue, controlled substances, and hazardous chemicals, activities and devices. If your project touches any of those — and far more projects do than students expect — the paperwork is not optional admin. It is a gate.

The obvious follow-up question is what counts as beginning. Reading papers, designing the study, buying equipment, building and testing your apparatus, and writing the research plan are all before the line. So, generally, is a pilot run purely to check that a rig works. The line is crossed the moment you record a number that will end up in your results. Write your approval date on the first page of your lab notebook and never put a data point above it.

When It Goes Wrong

The way this actually ends a project.

The pattern is depressingly consistent, so here it is as a sequence rather than a story about anyone in particular. A student has a good idea in September — say, whether screen use in the hour before bed predicts self-reported sleep quality. They write a questionnaire, put it in front of 140 classmates over two weeks in October, and get a real, interesting result. In November they build the analysis. In December they write it up. In January, filling in the entry forms, they reach the human participants section and discover that the study needed review board approval before the first questionnaire was handed out.

At that point there are two options, and neither is good. Option one is to declare it and accept that the data cannot be used. Option two is to re-run the entire study properly after approval, which needs roughly six weeks they do not have, and a second group of 140 people who have now already seen the questionnaire. Most students in this position quietly swap to a weaker project three weeks before the fair. Some do not declare it, which is worse: an undisclosed problem found by a judge is not a paperwork issue any more, it is an integrity issue.

I got lucky rather than being careful. The early version of my microplastic work happened in an outbuilding at home with oil-based ferrofluid, and for months there was nothing written down that resembled a risk assessment — no record of what was in the fluid, no plan for the used oil beyond a labelled jar on a shelf, no adult who had formally agreed to supervise. It was an adult asking me a blunt question about disposal that made me write any of it down. Nothing went wrong. It could have, and I would have had no answer for anyone.

The fix in both cases is the same and takes an afternoon: classify the project honestly before you start, and submit early enough that a committee can send it back once.

The Adults

Three different adults,
three different jobs.

The forms name several adult roles and students routinely assume they are the same person under different headings. Sometimes one person genuinely fills two of them — your chemistry teacher can be both sponsor and supervisor — but the responsibilities are distinct, and knowing which you are missing tells you who to go and ask.

If you have no obvious candidate for a Qualified Scientist, that is a solvable problem rather than a reason to abandon the project — it is mostly a matter of writing a lot of short, specific emails to people whose work is genuinely adjacent to yours. The guide on finding a mentor covers how to write those emails and how many to expect back.

Human Participants

Yes, your survey counts.

This is the trap. If a person is the source of your data, you are doing human participants research: questionnaires, taste tests, reaction-time apps, memory tasks, fitness measurements, interviews, and anything where you record what somebody did or said. Nobody is being injected with anything, so it feels exempt. It is not, and reviewers see this mistake more than any other.

There is a second layer for school projects: your participants are usually minors, and minors are a protected group. So a survey of your own year group needs written permission from a parent or guardian for each participant, the informed agreement of the participant themselves, and the agreement of the school as the institution letting you through the door. All three, before the first response.

A worked example

Suppose you want to test whether a two-minute walk improves reaction time. Your design: 60 participants aged 15 to 17, each doing a computer reaction-time task before and after either a two-minute walk or two minutes of sitting, randomly assigned, 10 trials each. Total participant time: about eight minutes. Risk: essentially the same as walking to a classroom, plus the mild embarrassment of a poor score.

That is a minimal-risk study, and the process is still not zero. You need a written research plan, a consent form for parents, an assent form for participants, a plan saying that names are never recorded and that each person is assigned a code at the desk, and the head teacher’s written agreement. Submit that package three weeks before you want to start. In my experience the committee will come back with one or two things — usually that your form does not say clearly enough that people can stop, or that you have asked for date of birth when age in years is all you need.

Note the sample size sitting in that plan. Deciding it in advance is not just an ethics requirement; it is also what stops you collecting until the result looks good, which is a much more serious problem. The statistics guide covers how to choose that number.

Animals

Don’t use vertebrates.
Genuinely.

I will be blunt, because the polite version of this advice has cost students entire years. At school level, vertebrate animal studies — mammals, birds, reptiles, amphibians, fish — carry the heaviest approval burden of any category, are usually restricted to registered research institutions with formal animal care oversight, and almost never buy you a scientific result that a simpler organism could not have given you. There is no bonus for using something with a spine.

The useful move is to work out what your question is actually about — a physiological response, a behaviour, a growth rate, a toxicity threshold — and then find the simplest system that shows it. This is a scientific improvement as well as an ethical one, because these systems give you far more replicates for the same effort.

Invertebrates are not a rules-free zone either. You are still expected to treat them humanely, avoid causing harm where the question does not require it, and have a plan for what happens to them afterwards — which should not be the sink. Some invertebrate groups are treated differently from others in some jurisdictions, so check rather than assume. Write the disposal plan into your research plan; a reviewer who sees it takes the rest of your document more seriously.

Chemicals & Microbes

BSL-1 is your ceiling.

Biosafety levels run from 1 to 4. For student research, level 1 is the practical maximum: known, characterised organisms that are not known to cause disease in healthy humans, handled in a supervised laboratory with prior approval. Anything above that needs a registered research institution, and no amount of enthusiasm substitutes.

The consequence catches almost everyone: culturing microbes you swabbed off a phone, a door handle, a shoe, raw chicken or the inside of someone’s mouth is not BSL-1 work, because you do not know what will grow. Unknown environmental and clinical samples are treated as the higher level by default. That popular project — swab five surfaces, incubate, count colonies — is exactly the one that gets stopped. If a supervised version is approved, the plates are never opened after incubation, they are counted through the lid, and they are sterilised or thoroughly disinfected before they go anywhere near a bin. Live cultures are not permitted on display at the fair, so your board shows photographs.

For chemicals, the deciding factor is not the name but the amount and the concentration. Household vinegar and glacial acetic acid are the same compound and a completely different risk. Read the manufacturer’s safety data sheet for everything you use — it is free, it is legally required to exist, and section 4 (first aid), section 7 (handling and storage) and section 13 (disposal) are written for exactly your situation. Learn the nine GHS pictograms; they take ten minutes and they tell you at a glance whether something is corrosive, an irritant, or a longer-term health hazard.

Some things are simply out of bounds for a school project or need a level of oversight you will not get: controlled substances, explosives and energetics, radioactive sources, mercury, high-power lasers, and mains-voltage electrical work. If a plan requires one of those, change the plan. It is not a rule invented to spoil your fun — it is that the failure mode is severe and irreversible.

Risk Assessment

One page, six columns,
one hour of your life.

A risk assessment is a table. For every hazardous material, piece of equipment or procedure, you write six things: the hazard, who could be harmed and how, how likely it is, the control measures that reduce it, the risk that remains after those controls, and what you do if it happens anyway. That is the whole document. Students imagine something bureaucratic; it is closer to a pre-flight check.

Here is a real one, written out in prose. The project measures how the rate of a reaction between magnesium ribbon and hydrochloric acid changes with acid concentration, at 0.5 M, 1.0 M and 2.0 M, five repeats per concentration, in a school lab.

Notice what makes that useful: numbers, named equipment standards, and the distance to the eyewash. A risk assessment that says “wear appropriate PPE and take care” is not a risk assessment; it is a sentence about wanting to be safe. Reviewers can tell the difference instantly, and so can a judge who asks you what you would do if you spilled it.

At The Bench

The basics, including the
kitchen-table ones.

A great many student projects happen at home, and home is where the assumptions live. These are the rules that matter most in practice, in rough order of how often they get broken.

Data Ethics

What you owe the people
who gave you data.

Anonymise at the point of collection, not afterwards. That means the name never enters the spreadsheet at all: each participant is handed a code at the desk, the code goes on the response sheet, and if you need to link a before and after measurement, the code is what links them. Collect the minimum that answers your question — age band rather than date of birth, year group rather than class, nothing at all where you can avoid it. Every extra field is a liability with no scientific payoff.

Store it properly. A spreadsheet shared with “anyone with the link” is published, whether you meant that or not. Keep the working file in an account only you can access, keep one backup somewhere else, and delete the raw responses when the project is finished unless you have told participants otherwise. If you are in Europe, data protection law applies to you as well as to companies, and your school will have a policy that someone can hand you.

Be careful about identifiability in what you publish, which is subtler than removing names. If your display board reports the results for “the two left-handed students in sixth year”, everyone in that school knows exactly who they are. Report groups large enough to hide in, and if a subgroup is too small to anonymise, describe it in words instead of breaking it out. Photographs of participants need their own separate consent — permission to take part in a study is not permission to appear on a poster.

And the ethics that has nothing to do with forms: report what you found. Do not delete the inconvenient data point, do not keep collecting until the difference appears, do not run six tests and present the one that worked. If you excluded a result, say so and say why, and ideally decide your exclusion rule before you see the data. Declare any substantial help you had, including from AI tools — judges are far more comfortable with a student who says what they did themselves than with one whose project is suspiciously seamless.

Before You Start

The order to do it in.

Work down this list before your first data point. Budget three weeks between submitting and starting, on the assumption that it comes back once — because it usually does, and a committee that asks you to clarify something is doing you a favour in February that would have been a disaster in April.

Two places worth having open while you do it: the official ISEF rules for the definitive answer on categories and forms, and Science Buddies for plain-English explanations of the same rules when the official wording defeats you. If you want somewhere sensible to archive your research plan, consent forms and raw data with a timestamp, the Open Science Framework is free and does exactly that — and a dated, public research plan is a genuinely impressive thing to be able to point a judge at.

Common Questions

The things students
ask too late.

Keep Going

Now go and do the
interesting part.

If you have not settled on a project yet, do this classification exercise while the idea is still soft — it is much easier to choose a design that avoids a heavy approval category than to rescue one that needs it. The Idea Lab will build a project around your interests, your equipment and your deadline, and it is free.