A short ungated path showing how DoOperator Education teaches: decision first, causal question second, optimization last.
Four-lesson preview
Each lesson below mirrors the full courses: a mental model, a practical exercise, and a standard for what good judgment sounds like.
A method is only useful after you know what decision it will inform. Before choosing an experiment, model, or policy optimizer, write the action you might take and the evidence that would change it.
Draft one sentence: If evidence shows X, we will do Y; if it shows Z, we will do W.
Replace vague claims like 'engagement improved' with a target comparison: what would have happened to this population under treatment versus control over a specific time horizon?
Convert a product idea into treatment, control, population, outcome, and time horizon.
Good systems do not optimize one number blindly. They track harm, validity, and operational failure: sample-ratio mismatch, support burden, complaints, latency, fairness, or escalation.
List one primary metric, two guardrails, and one condition that would stop launch.
Adaptive systems need a learning policy. Some decisions should use fixed experiments, some should use staged rollout, and some should wait for better logged data before automation.
Choose fixed experiment, staged rollout, observational analysis, or offline policy evaluation, then justify why.
Mini artifact
By the end of the sample, a learner should be able to write a compact brief that names the decision, estimand, metrics, guardrails, and next learning step.
What action is on the table?
What result would change the action?
What could make the evidence misleading?
What should the system learn next?
Continue learning
The full courses add notebooks, datasets, capstones, rubrics, worked examples, and certificates.
Design randomized experiments with estimands, power, metrics, and launch review.
Turn observational product data into criticizable causal designs and sensitivity checks.
Build logged-policy, off-policy evaluation, and safe rollout judgment.