# Synthetic lesson backfill This is invented teaching data, not a claim about a real challenge. After creating a toy lane, write a local evidence note and import one lesson: ```sh mkdir -p .runtime/lanes/sample-a printf '%s\n' 'Invented example: count-only control matched a toy model.' > .runtime/lanes/sample-a/control.md python3 harness/rh.py --root .runtime lessons add --lane sample-a \ --source backfill --tags synthetic,evaluation,shortcut \ --lesson 'In the invented example, a count-only control matched the model.' \ --evidence lanes/sample-a/control.md \ --applies-when 'Testing a synthetic label-frequency shortcut' python3 harness/rh.py --root .runtime lessons list ``` The generated runtime JSONL is local and ignored. Real evidence, human review, and an actual enforcing file are needed before promoting a real lesson.