Lessons
Grouped by unit. If you want them in the order to take them, the course page resolves the prerequisites for you.
30 lessons, about 403 minutes of work. Every one ends in a tool you operate on real data — not a screenshot of one. All free, no account needed to read.
Reading numbers honestly
Before any method, the skill everything else rests on: telling whether a number that moved actually means anything, and saying so out loud when it does not.
- Signal or noise: reading a number that movesFoundation · 14 min · 2 tools
- The average that describes nobodyFoundation · 13 min · 1 tool
- Work is a process, and somebody is waiting at the end of itFoundation · 12 min · 1 tool
- The difference between what hurts and what caused itFoundation · 11 min · 1 tool
- Two people, one measurement, different answersFoundation · 14 min · 1 tool
- The best worker, and why they are notFoundation · 13 min · 1 tool
Seeing waste
Learn to look at work the way an improvement practitioner does: as a flow of value interrupted by things nobody would pay for.
The method
DMAIC is a sequence, not a checklist. Learn what each phase produces, why the order is load-bearing, and how to state a problem without smuggling a solution into it.
- DMAIC: what each phase produces and why the order holdsWhite Belt · 12 min · 1 tool
- Writing a problem statement that does not contain its own answerWhite Belt · 12 min · 1 tool
- The hidden factory: why every step passes and the process failsWhite Belt · 13 min · 1 tool
- Mapping the gaps, not the boxesWhite Belt · 12 min · 1 tool
- Go and see: what a walk gives you that a report cannotWhite Belt · 12 min · 1 tool
- Scoping a project you can actually finishWhite Belt · 13 min · 1 tool
- PDCA: the Check is the part everybody skipsWhite Belt · 14 min · 1 tool
- 5S: why the tidy area keeps going backWhite Belt · 12 min · 1 tool
Finding causes
Most problems get solved by treating a symptom. These are the two cheapest tools for getting past that, and the judgement to know when each has gone far enough.
Control and capability
Two questions that get confused constantly: is the process behaving consistently, and is consistent good enough? They have different answers, different charts, and different consequences when you get them the wrong way round.
Comparing things honestly
Every change ends with the same question: is this difference real? The arithmetic is one line. Almost every failure in practice is in what the number gets taken to mean.
Relationships between measurements
A regression always returns a slope, an intercept and an R-squared — whether or not a straight line was ever the right shape. The residual plot is the part that tells you which.
More than two groups
A significant F says at least one group differs. It says nothing about which — and finding out by running every pair is how a study with no real differences produces one anyway.
Experiments that change everything at once
One factor at a time feels careful and cannot see interactions — the factors whose right setting depends on each other. A factorial runs every combination, and every run informs every effect.
Screening designs, and the price of a fraction
A full factorial doubles with every factor. A fraction buys breadth with confounding — and the bill is computable to the letter, before a single run is booked.
Finding the setting, not just the factor
Two levels cannot describe a curve. Adding the third finds where the gradient is zero — and the harder question is whether that point is a maximum, a minimum, or a saddle that is no optimum at all.
How many, and what it could show
The difference worth detecting, the sample size, alpha and power are one equation — so a study whose size nobody argued about has already settled one of them by accident. Solving it in whichever direction the project has left open is what stops a real improvement being written up as "no significant difference".