Rewriting Python Learning Protocol with Cold-Grill Diagnostics

The author runs an AI research lab that studies working with Claude Code. They discovered a failure mode in their learning protocol and pivoted to a 'build from scratch' approach to better understand Python's standard library.

Details

The author runs an AI research lab that studies working with Claude Code, not just using it. Their learning protocol involves two parallel tracks - one where Claude Opus is the coaching partner, and another QA-automation pipeline that tests the same operator pattern. They have a public framework called '.claude/' that codifies this approach. The lab deliberately runs failure modes to test the protocol. A recent cold-grill diagnostic on 19 Python concepts revealed that the author was struggling with standard library mechanisms, not just the execution model they had been focusing on. The author initially diagnosed this as 'threads not connected', but the partner corrected them to 'layers not walked'. This revealed the author's failure reflex of looking up solutions instead of truly understanding the concepts. To fix this, the author committed to a 'build from scratch' approach, recreating core Python components like dictionaries, lists, and JSON parsing from the ground up, with specific tooling and observations required at each step. This is meant to force the mechanism into the author's memory, rather than just completing exercises.