How to build a project

This page is the practical half of the pedagogy. It walks through six pieces of work every LBX entry needs: a question, a method, evidence, a prototype, a poster and an oral defence. Read it with the rubric open. Mentors will refer to these headings in written feedback.

Apply for the 2026 Season → See the wider field

Lead proposal card

Fill this before 30 November. Weak proposals are revised, not rejected. Applications remain open through 31 October 2026. Unique role of this site: multi-year inquiry and a global final.

Field	Write	Invented courtyard example
Observation	What you saw, where, on which days.	Midday wall in one Singapore school courtyard, ten school days in April.
Question	A sentence a neighbour could repeat.	Does this shade cloth change midday surface temperature on one wall, and by how many degrees?
Conditions	What you will hold still.	Same wall, same cheap thermometer, control patch without cloth.
Obtainable materials	What you can actually get by February.	Cloth, sewing kit, logbook, phone photographs of the mount.
How you will verify	A measurement, not a slogan.	Table of degrees. Ten days is not a season — say so.

1. Question

A usable question is narrow enough to finish by 30 April and wide enough that the answer is not already on the first page of a search. “How can we stop climate change?” is not a question you can own. “Does a low-cost shade cloth change midday temperature in one school courtyard in May, and by how much?” is. Write the question as a sentence a neighbour could repeat. Then write a second sentence: who is affected if you are right.

Test the question by trying to fake the answer in an afternoon. If you can, it is too small. If you cannot even name the first measurement, it is too large. Mentors are trained to push you toward a measurement, not toward a more fashionable topic. Changing the question after the proposal is allowed if you say so in the progress note.

2. Method

Method is the promise you make to a stranger: if they did what you did, they would see something like what you saw. Write it as a recipe — materials, steps, what you will count, what you will ignore. For software, that means a repository with a README that installs on a clean machine. For hardware, a bill of materials and a photograph of the bench. For a survey, the exact questions, the sampling frame and how you will store answers. For a chemistry or biology protocol, concentrations, times and what “it worked” looks like.

If people are involved — interviews, apps in a classroom, medical-adjacent devices — stop and write the ethics note at the proposal checkpoint. Consent is part of the method, not an appendix. You cannot recruit classmates on the day of the showcase and call it a study.

3. Evidence

Evidence is not a vibe. It is a table, a log, a graph, a set of commits, a video of the device failing in the same way twice. Collect more than you will print. Junior projects can live on a spreadsheet; Senior projects should say how error was estimated. When the evidence contradicts the hope you had in November, keep the evidence. The exploration score is built for this moment.

Cite the papers and pages you actually used. A bibliography copied from a language model without opening the files is a rules problem, not a style problem. Datasets you did not collect yourself must be named, licensed and described.

Build iteration log

Version the prototype. Judges like to hear the joint that cracked. This log is a demonstration of the Build column, not a fifth criterion.

Version	What failed	Test conditions	What you changed	Who did that bit
v1 mount	Seam opened in monsoon rain.	One courtyard wall, afternoon storm.	Moved the seam off the drip line.	Name the student who sewed it.
v2 mount	Cloth sagged at noon.	Same wall, no wind screen.	Added a fillet at the joint.	Name the student who cut the fillet.
v3 (progress checkpoint)	Still unknown: ten-day table not finished.	April school days, control patch.	Freeze the mount; collect the table.	Who logs the degrees each day.

4. Prototype

The prototype is the artefact the rubric calls “Build”. It has to run in the room. A rendering is not a prototype. A slide deck is not a prototype. A private GitHub repo the judges cannot clone is not a prototype. Plan the demo as if the network is down and the battery is at 12 percent. Bring a spare cable. If the thing is too large to fly, bring a section of it and a video of the rest, recorded before you travelled.

Version the prototype. Judges like to hear “this is the third chassis; the first two cracked at the joint, which is why the fillet is here.” That sentence is exploration made visible.

eXplore boundary table

Exploration is the X. Dead ends belong here. A limitations paragraph that names the confound you could not control beats a press-release conclusion.

Result you trust	Interpretation	A counter-example	Where it applies	Next test
On ten April days the shaded patch was cooler at midday on this wall.	The cloth changed that surface under those days.	A windy day, or a different wall, may not move.	One courtyard, April, this cloth. Not a climate claim.	Repeat in May or refuse to scale the sentence.

5. Poster

The poster is a map, not an essay. Title; question in one sentence; method in five lines; one figure that carries the result; limitations; next question. Print A1 portrait or export 2480 × 3508 px. Put your names and hub at the bottom, not as a watermark across the figure. Stand to the side of the poster during the public hour so visitors can read it. Do not block the figure with a laptop.

6. Oral defence

Seven minutes to present and demonstrate, eight minutes of questions. Memorise the opening sentence of the question, then talk to the artefact. Judges will ask what you would do with another month, what you would not do again, and who actually wrote the code or soldered the joint. “We” is fine; “the mentor did that bit” should be said if it is true. For teams, every member should be able to answer a method question. A member who only designed the slide theme will be obvious.

Practice once with someone who does not love the project. If they cannot repeat the question afterwards, the defence is not ready. Time the demo. A demo that overruns eats the question period, which is where the communication score is won.

If you are stuck, book office hours or re-read the four-criterion rubric. The FAQ covers AI tools, failed experiments and working with an existing mentor.

A worked path, without a named hero

Suppose a Junior team in the Singapore hub cares about heat in a courtyard. The question, after two mentor sessions, becomes: does a particular shade cloth change midday surface temperature on one wall, and by how many degrees, on ten school days in April. The method is a cheap thermometer, a logbook, a control patch of wall, and a photograph of the cloth’s mounting so someone else could hang it. The evidence is a table, not a slogan about climate. The prototype is the mounting, which failed twice in the monsoon and was sewn again. The poster has one figure. The defence starts with the question, shows the cloth, and admits that ten days is not a season. That project can score well. A slide deck titled “Saving the Earth with Nanotech” cannot, even if the animations are expensive.

Senior teams often overbuild the literature review and underbuild the artefact. Reverse that. Two pages of prior work that you actually read will beat eight pages of unread citations. A repository that clones in five minutes will beat a private notebook. A limitations paragraph that names the confound you could not control will beat a conclusion that sounds like a press release. Practice the defence with a teacher who is not in STEM. If they can repeat the question, you are close.

When you are stuck, do not add a feature. Write down the last measurement you trust and the next one you do not. Bring that pair of sentences to the mentor. That is Lead and eXplore in one meeting. Build will follow, or it will not, and either way you will know before April.

Common ways a project dies in March

The first is scope: a team keeps adding sensors because the current one is embarrassing. Stop. Document the embarrassment; that is exploration. The second is authorship: one member holds the repository password and the others hold the slide theme. Split the password in February, not the night before the flight. The third is ethics delayed: a survey designed in April cannot be run with consent in time. If people are your evidence, they were already in the November proposal. The fourth is travel magic: the prototype that “will be finished in the hotel.” It will not. Freeze a week early and demonstrate what exists.

Office hours exist for statistics, for soldering questions, and for the sentence you cannot write. They are not a rewrite service. Bring the last measurement you trust. Leave with the next one. Repeat until 30 April. That is the whole method of the second half of the season, and it is more reliable than inspiration.