Order-of-Magnitude Optimizations — Week 2 CheckPruning, incremental economics, sketches, and layout debt. Every question is answerable from the Week 2 brief.

Snowflake-style warehouses have no indexes. What actually limits how much data a filtered query reads?

A dbt model uses SELECT * inside a CTE, but downstream only two columns are consumed. What does the engine actually read?

You added is_incremental() to a model and the bill didn't move. What do you check FIRST?

A file-compaction service scores partition health with mean SQUARED error of file sizes against target, rather than mean absolute error. Why?

On a merge-on-read table, why do equality deletes degrade reads so much more than positional deletes?

You inherit 400 pipelines and a 30% cost-cut mandate. Per the effort-ranked triage, what comes first?