The order a query means, and the plan DuckDB actually runs, read from real output.
Two A4 pages on what happens between typing a query and getting rows back, on HealSync's tables from Explain Your Query. Every result was run, and every plan shown is DuckDB's own EXPLAIN output, trimmed.
It starts with written order against run order and what follows from it: which databases let a SELECT alias into WHERE, GROUP BY or HAVING, an alias that shares a column's name, WHERE against FILTER when a group would be empty, and why a filter cannot move across a window. Then how to read EXPLAIN and EXPLAIN ANALYZE, measured on a million-row copy of the claims: which filters let DuckDB skip blocks of rows and which do not, what SELECT * reads in a column store, what LIMIT changes in a plan, and what DuckDB did with a CTE used once, used twice and written as a subquery.
It ends with what an index does in a row-store database such as PostgreSQL, SQL Server or MySQL: keys, column order, covering indexes and their cost, stating only documented behaviour.
PDF to print, and an HTML version to read on screen. DuckDB syntax, with notes for PostgreSQL, SQL Server and MySQL.