Teach uncertainty with an engine students can open.
Probability, calibration, evidence discipline and reproducibility — taught not from slides but from a live, glass-box decision engine that runs free in every student’s browser and says “insufficient evidence” instead of guessing.
Why teach with it
Most people leave education able to calculate a probability and unable to use one. LEXUN is a working decision engine whose every design choice is a lesson in applied uncertainty — and because it is a glass box, students can open it.
Bands, not points
The engine answers "60–70%", never "67.3%" — a live demonstration of what precision the evidence does and does not support.
Refusal as rigour
Thin evidence gets "insufficient evidence", not a guess. Students see a system that would rather be silent than wrong.
Reproducibility
Seeded simulation means every student, on every laptop, gets the identical result from identical inputs — the scientific method, enforced by code.
Open source of truth
The models are auditable JavaScript with stated weaknesses and version numbers. Nothing is "the algorithm says so".
Three classroom exercises that work today
1 · Reproduce a published result
Give the class the reproducibility ID from the worked example on Reproduce a result. Every student reruns the seeded simulation and verifies they get the same band, then changes one input and explains why the band moved. The public forecast register teaches the other half: its forecasts are editorial judgements with no seed to rerun, frozen with resolution dates already on official calendars, so a class can read a prediction before reality answers it, recompute its receipt, and come back to score it. Verification before trust — practised, not preached.
2 · Facts versus assumptions
Students frame a real decision — a gap year, a first car — and must classify every input as fact, assumption or unknown before the engine will run. The sensitivity analysis then shows which assumption their whole answer secretly rests on.
Frame a decision → LIVE
3 · Forecast, freeze, score
Each student makes a forecast with a frozen resolution criterion and a due date. When the date arrives, outcomes are reported and Brier-scored — the class builds its own calibration record and discovers, with numbers, whether their 80% really meant 80%.
Teacher note: this fits a term exactly — freeze forecasts in September, score them in December. Fits A-level maths (probability, expectation) and economics (forecasting, uncertainty), and intro university statistics (scoring rules, calibration). The loop — predict, freeze, wait, score — is the method teaching itself.
The ninety-second calibration test
A classroom starter that teaches the central lesson of this whole page in one round: being right and knowing how right you are are different skills. Ten true-or-false statements; for each, pick an answer and how confident you are. The score is a Brier score — the same rule that scores LEXUN’s public forecasts. Everything runs in this page; nothing you answer is stored or sent anywhere.
Ten questions is a taster, not a measurement — a usable calibration record needs far more answers, which is why LEXUN refuses to display an accuracy percentage anywhere before ten resolved real outcomes exist in a domain. Teacher note: run it twice, once before and once after teaching what the confidence levels mean; the improvement is usually visible immediately.
The teacher one-pager
LEXUN in the classroom — one page
What it is: a free, glass-box decision engine at lexun.co.uk/platform. It turns a written decision into declared inputs, refuses to run until facts and assumptions are separated, returns a probability band with its drivers, and freezes forecasts so they can be scored against reality.
Setup: none. Browser tabs. No accounts, no installs, no student email addresses; the engine does not transmit what students enter. Use fictional scenarios, keep the optional Copilot out of lessons, and run without saving on shared machines.
Starter (10 min): the calibration test on lexun.co.uk/education — ten true/false questions with confidence levels, Brier-scored instantly.
Lesson 1 — Reproduce a result: students re-run a seeded simulation from lexun.co.uk/reproduce and verify the identical band, then move one input. Concept: reproducibility, sensitivity.
Lesson 2 — Facts vs assumptions: frame a real decision (gap year, first car); classify every input before the engine runs. Concept: evidence discipline.
Lesson 3 — Predict, freeze, wait, score: forecasts with frozen criteria in September, Brier scores in December. Concept: calibration. The public register at lexun.co.uk/forecast shows the same loop running in public, misses included.
Fits A-level maths and economics, and university intro statistics. Free is architectural, not promotional: the engine computes on the student’s device. Questions: hello@lexun.co.uk
Printing this page gives you the one-pager box on its own sheet — or link colleagues straight to lexun.co.uk/education#one-pager.
What it costs a classroom
Nothing, and that is architectural. The engine runs in the browser on any laptop or school machine — nothing to install, no accounts to provision, no student email addresses collected, and the engine keeps every figure on the device; the optional Copilot, the one feature that sends typed text to an AI provider, is not needed for any exercise here. A class of thirty needs thirty browser tabs.
Privacy in the classroom. The core simulation engine runs locally and does not transmit students’ decision inputs. The optional Copilot is separate: it is connected on this deployment, and it sends the text typed into it to an external AI processor (Anthropic) through a LEXUN function, only when someone presses its button. When you teach:
- Use fictional scenarios — the worked examples, the templates and the public forecast register — rather than students’ own or their families’ finances. Every exercise above works that way.
- Do not use the Copilot with student or other personal information; no exercise here needs it.
- On shared computers, run decisions without saving. Running needs no Key and no account, and the decision is not written to the browser unless someone saves it. A few small settings the site keeps (the theme, for instance) are listed with their purpose on the cookies page.
- If a student does save, the record stays in that browser, encrypted under the passcode they chose, until it is deleted. To clear a machine, use Delete all data in the workspace’s Settings, or the erase control on the cookies page; both act only on that browser.
- No name, email address or account is asked for to run a decision.
LEXUN claims no compliance with any schools framework. The privacy notice applies to students exactly as it does to everyone, and your school’s own data-protection policy decides what is appropriate.
Honest limits
LEXUN is a decision engine, not courseware. There is no LMS integration, no gradebook, no lesson-plan library and no teacher dashboard today — and this page will say so until that changes. The three models are UK-flavoured (sterling, UK statistics as reference classes) and cover runway, career and purchase decisions — they are a laboratory for uncertainty, not a statistics curriculum. Each model publishes its own stated weaknesses, which make good seminar material in themselves.
Run a pilot
If you teach statistics, economics, business, computing or critical thinking and want to trial LEXUN with a class, tell us the shape of the course. We are a small team; pilots get a direct line to the people who wrote the models.