Easy: Re-run 01-prompt-suite-bol-extraction.ipynb with a different model (set OPENAI_MODEL or OPENROUTER_MODEL) and report the version-3 score delta.
Standard: Add a 4th prompt version with chain-of-thought ("think step by step about the cargo block first"). Compare its accuracy and token cost against version 3.
Stretch: In 02-context-engineering.ipynb, add a tiered eviction policy: label each budget claimant as pinned / compressible / disposable, then re-run the worksheet with a 3× longer document and show what gets cut first.
Portfolio: Save the prompt suite as projects/bol_extractor.py with a JSON eval log (version → score → note) and a gate that fails below 90% per-field accuracy.
Easy: Set OPENAI_MODEL (or OPENROUTER_MODEL) before running; re-run only version 3 and diff its OVERALL row against the default model's score.
Standard: Add a version == 4 branch to make_prompt that prepends a "reason about the cargo block first" line; compare its accuracy and its token cost (the CoT text is extra input tokens).
Stretch: Give each worksheet claimant a tier label (pinned/compressible/disposable) and, when the document triples, evict in that order before re-measuring the total against the ceiling.
Portfolio: Wrap the three make_prompt versions and the grader in a function, dump \{version: score\} to JSON, and sys.exit(1) when the best score < 0.90.