Crash course

Before any tooling: the vocabulary, the mental model of a dataset, and the single most valuable skill in the job — converting "can you pull some numbers" into a question that has a defensible answer.

1. 01.01 What a data team is actually for 3 min read
2. 01.02 Rows, grain, and the shape of a dataset 3 min read
3. 01.03 Turning a request into a question 4 min read
4. 01.04 A working setup that will not embarrass you 3 min read

Build a mental model of a pipeline that survives contact with reality: late data, duplicate deliveries, schema drift, and the 3am rerun.

1. 02.01 What a pipeline is really doing 4 min read
2. 02.02 Batch, streaming, and the honest difference 4 min read
3. 02.03 Idempotency, backfills, and yesterday arriving twice 4 min read
4. 02.04 Schema drift, and the contract that prevents it 3 min read

Learn to design tables an analyst can use without asking anyone a question, and understand the storage layer well enough to know why a query costs what it costs.

1. 03.01 Facts, dimensions, and picking a grain 4 min read
2. 03.02 Slowly changing dimensions, and the "as of when" problem 3 min read
3. 03.03 Why your query costs what it costs 4 min read
4. 03.04 One metric, one definition 4 min read

Get past "I can write a GROUP BY" to the SQL that appears in real work: window functions, careful joins, and the ability to explain why a query is slow.

1. 04.01 Joins that do not fan out 3 min read
2. 04.02 Window functions, properly 3 min read
3. 04.03 Writing SQL people can read, and reading the plan 3 min read
4. 04.04 The patterns that keep coming up 2 min read

Move from "the pipeline ran" to "the numbers are right", with tests, contracts, and a reconciliation habit.

1. 05.01 Testing data like you test code 4 min read
2. 05.02 The anatomy of a silent failure 4 min read
3. 05.03 The first hour with an unfamiliar dataset 3 min read

Build the statistical intuition the job requires — which is less about tests and more about knowing how much you should trust a number.

1. 06.01 Describing a column without lying 4 min read
2. 06.02 Uncertainty, sampling, and how much to trust a number 4 min read
3. 06.03 The five tests that cover most questions 4 min read
4. 06.04 Correlation, confounding, and Simpson's paradox 4 min read

Learn to run experiments that are worth running, and to make careful causal claims from observational data when an experiment is impossible.

1. 07.01 Designing an experiment that can succeed 4 min read
2. 07.02 Reading a result without fooling yourself 4 min read
3. 07.03 Causal claims when you cannot randomise 4 min read

Build predictive models whose offline numbers survive contact with production — which mostly means getting validation and leakage right.

1. 08.01 Frame the problem, then beat a stupid baseline 5 min read
2. 08.02 Validation that matches deployment 3 min read
3. 08.03 Features, and explaining what the model did 4 min read
4. 08.04 When not to build a model 4 min read

Make charts that answer the question they were built for, and communicate a result to someone who will not read your notebook.

1. 09.01 Choosing the form before the colours 4 min read
2. 09.02 Colour is an encoding, not decoration 4 min read
3. 09.03 Communicating a result to someone who will not read your notebook 4 min read

Understand what changes when a piece of analysis becomes a thing that runs unattended, and what you owe the person who gets paged.

1. 10.01 Orchestration, and what "it runs every night" costs 4 min read
2. 10.02 Monitoring data, not just jobs 3 min read
3. 10.03 Owning something that runs without you 3 min read