Worked examples · From engagements, shown with permission and identity removed · EIS-001
Catching a student before the term does
Early intervention signals across a synthetic cohort. Education on the surface, and the shape is any early warning problem where acting late costs more than acting wrong.
OpenSynthetic dataFixed seed
This run
184 students · 12 weeks · fixed seed
What the simulation shows
Every figure here is generated from a fixed seed, so the numbers on this page are the numbers in the simulation.
The entry
The problem
Attendance and submission data exist from week one, and the intervention happens after results are published, when nothing can be changed.
Why it matters
The cost of acting late is a lost year. The cost of acting wrong is a conversation. Those are not comparable, and a threshold set as though they were will always be set too high.
The approach
Combine attendance, submission and engagement into one signal per student per week, and mark the week the signal crosses a threshold that still leaves time to act.
Judgement calls
The threshold is deliberately set to over refer rather than under refer, and the page says so. A model tuned for precision on this problem is a model optimised to do nothing.
Where else it applies
Any early warning problem with an asymmetric cost of error: churn, credit, safeguarding, clinical deterioration, project slippage.
The data
Entirely synthetic, generated from a fixed seed. No student record from any institution was used, seen or derived from.
Open the simulation
It asks for your name and email, then opens. The page carries a watermark with the name of whoever opened it.