Theory of change
The logic model, written so you can attack it
Every link here is a claim you can test now, except the last one, which cannot be tested until 2036.
- Twelve labs and 240 free written project steps
- Trained human assessors and a certification panel
- AI marking at about US$2 per learner yearly
- Grant funding for seats, translation and assessor training
- Student picks a free project in any lab
- Interviews one expert, three affected people, three peers
- AI marks artifact, process and defence out of 20
- Human panel certifies or returns written feedback
- Projects submitted, assessed and returned within two weeks
- Learners certified at Level 1, 2 or 3
- Ten relationships logged and named per project
- A portfolio with a public share link
- Students return for a second project, different lab
- Evidence of capability a stranger can check
- A named network of experts, peers and reviewers
- Teachers certified, and they keep the certification
- Measurable options at eighteen beyond sending out CVs
- Work arrives before applications do, or it does not
- Checkable publicly in 2036, including if it disappoints
Read left to right. Everything left of Outcomes is delivered today; Outcomes and Impact are what a grant is meant to buy, and both are stated as targets.
Unit economics
What a grant buys per learner per year
Four cost lines make up a certified learner. Only one of them is expensive, and it is expensive on purpose.
Measured in US$ per learner per year
These are planned unit costs from the operating model, not audited accounts.
See the numbers as a table
| Line | US$ per learner per year | Note |
|---|---|---|
| Human certification | 12 | Panel decision, does not scale linearly |
| Platform, storage and support share | 8 | Hosting, share links, help |
| AI marking and defence questioning | 2 | Marked out of 20 |
| Parent reports | 2 | Written back to the family |
What money changes
Four things a grant unlocks that free content cannot
The content is already published and permanently free. These four are people, and people have to be paid.
Human certification
AI already marks every project. A human panel decides whether it counts, at about US$12 per learner per year. It is deliberately the part that does not scale.
Assessor training
Teacher Certification runs at US$2,320 per cohort of 30: two days onsite plus four weeks of guided practice. The teacher keeps the certification.
Sponsored seats
US$60 covers EAD Plus and certification for one family that cannot pay. The rule holds everywhere: the gate is the assessment, never the fee.
Translation and reach
Getting the No-Tech route, which needs only paper and conversations, into schools with no electricity and no budget. Languages: [TRANSLATION LANGUAGES — CONFIRM].
Measurement
What we count, how we count it, and when you see it
The first row is the one metric above all others. The last row is the one we cannot report yet.
| Indicator | How it is measured | Reported |
|---|---|---|
| Second project, different lab | Share of students who finish a first free project and start a second one in another industry | Quarterly, by cohort |
| Projects submitted | Count of completed assessment templates received through the submission route | Monthly |
| Projects assessed | Count returned with written feedback on every section, inside the two-week window | Monthly |
| Learners certified by level | Panel decisions logged at Foundation, Level 1, Level 2 and Level 3 | Quarterly |
| Relationships logged | Ten per project: 1 expert, 3 people affected, 3 peer reviewers, 3 shares that replied | Per project, at assessment |
| Longitudinal outcomes | Not measured yet. The 2026 cohort can only be followed to 2036 | [BASELINE STUDY — CONFIRM] |
How it runs
From signed agreement to independent verification
Five stages. You choose the evaluator at the last one, and we do not see the sample before they draw it.
- 1Agreement and scopeWe fix the number of sponsored seats, the labs, the regions and the reporting dates. Unit costs are held at the figures on this page.
- 2Assessors trainedA cohort of 30 teachers takes two days onsite and four weeks of guided practice. They then mark alongside a certified panel member.
- 3Students do the workThree or four free projects a year alongside school, in different industries. High-Tech or No-Tech, identical curriculum either way.
- 4Assessment and certificationAI marks the artifact, the process and the defence out of 20. A human panel certifies. Portfolios stay marked unverified until it does.
- 5Reporting and verificationYou receive the counts and the second-project rate. An evaluator you choose may sample submissions, re-mark them and call the logged contacts.
Safeguarding and data protection
What is in place, and what a grant would formalise
The rules on the left are written into the projects and the assessment template today. The right side is honest about gaps.
What is already built in
- Every under-18 submission carries parent or guardian consent.
- The assessment template asks for names only, and explicitly forbids phone numbers and addresses.
- Projects instruct students never to record complaints about named individuals.
- Fieldwork rules are plain: never go alone, never enter a home.
- Portfolios stay marked unverified until a human panel certifies them.
- Community posting opens only after one completed project, so no anonymous adults arrive first.
What a grant would formalise
- [SAFEGUARDING POLICY — CONFIRM] written, published and version-controlled.
- [ASSESSOR VETTING PROCEDURE — CONFIRM] before any adult marks a child's work.
- [DATA RETENTION SCHEDULE — CONFIRM] for submissions, marks and public share links.
- [INCIDENT REPORTING ROUTE — CONFIRM] with one named responsible person.
- Safeguarding added to the assessor cohort as an assessed module, not a briefing.
What would prove this model wrong
There is no longitudinal outcome data yet, and conversion from free to paid is unproven at scale. Human certification does not scale linearly, and projects P03 to P12 exist as titles and outlines only, because just the opening project in each lab is written out step by step so far. If most students who finish a free project never start a second one in a different lab, the discovery model has failed, and the 2036 cohort data will settle the ten-year claim publicly, including if it disappoints.
Targets, not results
What we are aiming at, and when you can check it
Every number below is a target from the operating plan. None of them has been achieved.
- Year 1 · to July 2027Target: the first cohort at scale20,000 students completing at least one free project, 25 paying campuses, 600 paying families, 100 professionals certified and 1 sponsored lab, at roughly US$13,000 to US$16,000 per month by month 12.
- Year 5 · by 2031Target: the levels get recognised150,000 free learners, 500 campuses, 6 sponsored labs, 3 institutions formally recognising EAD levels and US$1.8M to US$3M ARR. Level 2 and 3 learners write projects for those below them.
- Year 10 · by 2036Target: the claim gets testedThe 2026 cohort turns twenty holding 50+ certified projects and a network of about a thousand people, at the age their classmates begin sending CVs. Either work arrives first or it does not.
Free download
The Grant Pack
One email, four documents, no audited accounts and no evaluation results, because neither exists yet.
- 01A two-page concept note: the problem, the model, and exactly what a grant buys
- 02The logframe: indicators, means of verification and reporting frequency
- 03A budget template carrying the real unit costs of US$12, US$8, US$2 and US$2 per learner per year
- 04The safeguarding and consent summary, with every gap marked [SAFEGUARDING POLICY — CONFIRM]
Here it is
We have emailed these links as well, so they are easy to find later.
Before you write back
Questions we get asked
If yours is not here, write to us. The answer usually ends up on this page.