Pedagogy Lab (Control of Error)¶
This worksheet shows how “practice” (coherence) increases evidence (and, with I_s > 0, utility), how a prepared environment fixes constraint loops, and how to read a routing certificate.
1) Run the Demo¶
python examples/pedagogy_control_of_error.py
Observe:
Before practice: decision, utility, distance, evidence.
After practice: evidence increases and (with I_s > 0) utility increases.
Prepared environment: region JP is infeasible; region US is feasible.
2) Tweak beta_s (coherence weight)¶
Edit configs/router_defaults.yaml and adjust beta_s (coherence weight). Re-run the demo and observe how evidence affects utility.
3) Explain a Flip¶
Change the prompt (e.g., add punctuation or replace a word). If the decision flips, use the certificate fields (distance/evidence/feasible) to explain why.
For a single certificate JSON, you can render a quick Markdown card:
python tools/visualize_certificate.py --input reports/certificates_demo.json
4) Classroom Pack¶
Generate a small, self-contained pack for workshops:
python scripts/generate_classroom_pack.py
This writes artifacts/pedagogy_pack.zip with:
Demo script, lab worksheet, a sample certificate JSONL, and a tiny prompt set.