Leader track 🧭 · six sessions, no code
For anyone handed a plan and asked to approve it. Why the obvious allocation is usually wrong, what a constraint is actually worth and for how long, what "optimal" quietly assumed, and the five questions to ask about a plan you did not produce.
Prediction is half the question
A forecast says what will happen. It does not say what to do, and the gap is where most analytics stops.
📉 Leader session 2The highest margin product nobody makes
Show earns 71 a tonne and appears in no optimal plan. Margin ranks products; the mill allocates hours.
🏷️ Leader session 3What a constraint is worth
One resource is worth 33.79 an hour, one is worth nothing, and the first number expires at +0.46.
🎲 Leader session 4The plan that works on average
Promises 4,656. Delivers about 4,325, and sells out a quarter of the time.
⚖️ Leader session 5When optimisation beats a model
Which problems are really allocation wearing a forecast, and where an optimiser is confidently wrong.
🔍 Leader session 6Reading someone else's optimum
Five questions for a plan you did not produce, each with the answer that would change your mind.
Practitioner track 🏗️ · ten sessions, hands on
For the people who build the model. Writing the problem down, the feasible region, simplex, shadow prices and their range, integers, formulation traps, Monte Carlo, robust planning, a solver that was wrong twice, and a capstone bench.
Writing a problem down
Decision variables, an objective, and the constraints. Most of the work is here.
📐 Practitioner session 2The feasible region
Why the optimum sits on a corner, and what that buys you.
⚙️ Practitioner session 3Simplex, mechanically
Pivoting, slack variables, and reading a tableau without fear.
📏 Practitioner session 4Shadow prices and their range
The dual, and measuring the exact point where the rate stops holding.
🔢 Practitioner session 5Integers change the answer
Rounding costs 60.00 and points the wrong way: 30 against a true optimum of 33.
🪤 Practitioner session 6Modelling choices that bite
Units, big-M, and the formulations where the model is the bug.
🎰 Practitioner session 7Monte Carlo over the plan
Draw the demand, re-run the plan, and read the percentiles instead of the mean.
🛡️ Practitioner session 8Robust and stochastic, in outline
Planning for a distribution rather than for its average.
💥 Practitioner session 9The solver that was wrong twice
It passed three hand-checked cases. A cross-check over 1,100 problems found two faults.
🏗️ Practitioner session 10The optimisation bench
Solve, round, branch, price and simulate, all live in the page.
The whole thing on one page
Every idea in the course and the session that owns it. Click any node to jump there.
Three ways through
Everyone starts at session 1 of their track. These are the shortest useful routes if you cannot take all sixteen.
What is real here
The solver is a simplex implementation written from the method, with branch and bound over its relaxation. Every plan on every page is computed, not stored.
| The solver | Carries three problems whose optima can be checked by hand, and runs them on load |
| And it was wrong | Passing those three was not enough. A cross-check over 1,100 generated problems found two faults in opposite directions, one in the solver and one in the checker |
| Every shadow price | Quoted with the range it holds over, because 33.79 an hour stops being true after 0.46 of an hour |
| Every plan | Run against drawn demand as well as average demand, because those give different answers |