Checkpoint
One last thing before we move on. Same surface as a write step — but the lesson doesn't complete until this passes.
Assemble the full rubric with its edges: (1) tabular WITH text_columns → "mixed: ladder for numeric + embeddings for text"; (2) spatial images → "CNN - locality + weight sharing fit"; (3) sequences with long-range deps → "transformer - attention spans the context"; (4) plain tabular → "ch.39 ladder first"; (5) anything else → "REFUSE: unknown shape - describe the data before picking". Use .get() for the optional keys. Print all five cases as f"{name:<15} -> {rec}". Specific rules before general ones; refusal over glibness.