01 Source baseline — instructor notes

Timing

Budget 20 minutes. Call time at 12 minutes for tool inspection so every pair reaches the source summary before design review.

Talk track

Have learners inspect the source OLAP pack, then click Discover & Design Schema. Explain that the button injects a source-specific prompt; it is not a link to a separate script. In expanded tool activity, require databricks-source calls—list_tables, run_select_query, and describe_table—rather than unrestricted Python discovery.

Before target DDL, the chat must show live table counts, semi-structured types, timestamp distinctions, liquid clustering, deletion vectors, generated columns, and the optional materialized view. The copied conversation link is the baseline record.

Common failures

• The query editor contains ClickHouse rewrites instead of the original source workload.
• The agent lists metadata but never obtains live counts.
• Delta metadata is mistaken for COUNT(*).
• Tool activity is collapsed, so learners cannot distinguish MCP discovery from Python.
• A sample value is accepted when a live source call failed.

Reset steps

Restore the source query pack through Edit · OLAP, start a clean conversation if needed, and re-fire Step 1. Ask for only the missing live fact in chat; do not move the group to a parallel collector. Stop before DDL approval until the baseline is complete.