SQL FOR DATA ANALYSIS / MODULE 03

This package contains only fictional training data. No accounts, secrets or live client information.
Open northstar_sql_practice.sqlite in a local SQLite-compatible editor. Work on a copy. The website has no arbitrary SQL executor and does not need your database credentials.
Alternatively, create a NEW EMPTY local SQLite database and run setup_sqlite.sql. In the SQLite shell, .read setup_sqlite.sql is a shell command; other editors have their own script-run control. Never use your website/hosting database. The setup has no DROP commands and will fail on conflicting existing tables.
Run SELECT sqlite_version(); Window exercises need SQLite >=3.25.0. The solutions were executed using SQLite 3.54.0. Other SQL dialects are not claimed tested.

Four tables: orders (12 verified orders); order_items (24 synthetic teaching lines); channels (web, store, partner; partner has no orders); raw_orders (13 imperfect source rows). Original order values match Modules 1 and 2, now stored in integer CENTS. Divide by 100.0 to present EUR. Additional item splits and the empty partner dimension are teaching additions, not original source observations.

Start with practice_queries.sql. Its statements are valid read-only STARTERS, not completed answers. Complete one task at a time. worked_solutions.sql has the 14 final SELECT statements. expected_results.json records output columns, order and values; null is SQL NULL, not zero. CSVs are alternative inspection/export files. No import is needed when opening the ready SQLite file.

Reference: completed sales Feb EUR500 / Mar EUR600; completed counts4/4; AOV125/150; March growth20%; cumulative completed sales1100. The source duplicate is M004. M006 has an invalid date and missing channel; M005 has a negative original value. Verified orders retain corrected source-backed values.

Period coverage is full Feb/Mar2026 by case assumption. Statuses are final snapshots, not refund events. Values are not profit or accounting revenue. No customer, campaign, advertising cost or causal evidence is included.
Each query has an ORDER BY where output ordering matters. Monetary sums stay in cents until presentation. Keep source, scripts, engine version and an evidence note together. Local lesson knowledge checks give practice feedback; they do not execute or grade arbitrary submitted SQL.