01
Rules settle outcomes
Agents propose what they do, and the engine resolves it by rule. No model call changes the school's true state.
Rehearse school decisions before real students live with them
A complete school, played by AI agents. You lead it.
Synthetic School runs a full school year. Every member of staff and every family is an AI agent with a role and a memory. Every week ends at a save point: return to one with a different decision and watch both versions of the school play out.
The school you can rewind
Two copies of one school
They share every random draw, so they stay identical, student for student, until the week you choose. After that, every difference between them comes from your decision.
{{ seedLabel }}
{{ when }}
{{ status }}
Still under 90%: A {{ underA }} · B {{ underB }}
Assumed effect of support
{{ effectNote }}
{{ srSummary }}
Illustrative simulation running in your browser, with 1,000 synthetic students in each school. Each dot is a student, and a brighter dot means better attendance so far. Hollow dots are under 90%, and a dot shrinks on any day its student is absent. Support cuts each supported student's daily chance of absence by the setting above, phased in over three weeks. The settings come from published studies; the three-week phase-in is an assumption.
The school you can rewind
Two copies of one school
They share every random draw, so they stay identical, student for student, until the week you choose. After that, every difference between them comes from your decision.
Seed 7041
Week 1 · school day 1
Both schools are identical, student for student.
Assumed effect of support
Central: about what mentoring achieved in a randomised trial.
Animated illustration of two identical synthetic schools of 1,000 students each, with a decision at the end of week 7 and an assumed effect of 2.5 points. Both schools are identical, student for student.
Illustrative simulation running in your browser, with 1,000 synthetic students in each school. Each dot is a student, and a brighter dot means better attendance so far. Hollow dots are under 90%, and a dot shrinks on any day its student is absent. Support cuts each supported student's daily chance of absence by the setting above, phased in over three weeks. The settings come from published studies; the three-week phase-in is an assumption.
A working school made of AI agents
The engine decides. The agents narrate.
Agents give the school its voices, and the engine's rules give it consequences you can trace back to their cause.
Built for every school system
One engine. A pack for every school system.
Every school shares a skeleton: a timetable, attendance records, students with additional needs, staff, a budget, families and someone the school leader answers to. A pack holds what differs between systems, and every fact in it carries its source.
What a pack holds
- School calendar
- Local job titles
- Attendance codes
- Child protection procedure
- Suspension and expulsion rules
- Inspection or accountability framework
- Leadership standard
- Funding rules
- Exams and qualifications
- Published statistics
- Personas from public sources
Where a system publishes no attendance figures, a design partner supplies the target and the pack marks it as partner data.
- A public school system · Somewhere HighIn build
- International schools under a regulator · Somewhere InternationalIn build
- Five more systems · on four continentsSources mapped
Each new pack is reviewed by a serving school leader from that system.
Test it here before it reaches a real school
One school, three uses.
For school leaders and the people who train them
Leadership training
Run a school for a term or a year, then learn from a debrief that maps your choices onto your system's leadership standard. Leadership programmes and school groups run it in cohorts.
How a training cohort runsFor companies building AI for schools
A test harness
Put your agent to work in a school with no real student data, through a REST API or an MCP server. Every action is logged, and the run report shows what followed.
What the test harness reportsFor school groups and education systems
A policy lab
Run a policy across hundreds of synthetic schools and compare the spread of results before you try it on real ones.
How the policy lab worksEvidence so far
Calibrated against published figures.
A working prototype of the engine has been calibrated in two school systems so far, to five published figures in all. On 200 fresh synthetic schools in each, the fit holds within 0.2 points of every published figure. Each new pack is calibrated the same way, against its own system's figures.
System A · secondary schools, 2024–25
Attendance, students from the most deprived fifth of areas
Published
84.2%
Engine
84.4%
Attendance, students from the least deprived fifth of areas
Published
92.4%
Engine
92.4%
Persistent or chronic absence, missing 10% or more
Published
37.4%
Engine
37.4%
System B · state-funded secondary schools, 2024–25
Attendance, all students
Published
91.6%
Engine
91.4%
Persistent or chronic absence, missing 10% or more
Published
23.4%
Engine
23.6%
About 0.02 s to run a year of attendance in a 1,000-student school
About 5 s to run the same year in 300 schools
Published figures: each system's official attendance statistics for 2024–25, listed in full on the Evidence page. Engine figures: the mean of 200 synthetic schools generated from new random seeds after fitting. The prototype covers attendance only, and the full engine is in build.
Safety
No real child is ever in it.
Synthetic only
Every school and every student in it is synthetic. No import path accepts real student or staff records.
Formative only
Nothing leaves as a score or a ranking. Contracts bar any use of play records in selecting or appraising a school leader.
Reviewed storylines
Child protection storylines come from a reviewed library and follow the system's procedure. They reach you only as reports.
Your record, your call
The trainee owns the play record, stored in a UK or EU region, and decides who sees it.
Run the first cohorts with us.
We are looking for leadership programmes and school groups to run the first pilots, free or at cost while Synthetic School is validated. A pilot in a new system starts with building that system's pack together.