What does a "skill checkpoint" actually test on these paths?

Each checkpoint is a 30-minute live exercise where you ship a working artifact without Claude Code, then a 30-minute review where you ship the same artifact with Claude Code. Example checkpoint at week 2 (backend path): "Add a discount_reason attribute to sales_order via db_schema.xml, expose it on OrderInterface, save it from a plugin around QuoteManagement::placeOrder." If the without-AI run is > 4× slower than the with-AI run, you have not internalised the pattern — loop back to the lesson.