This folder contains fixed evaluator examples and staged search procedures.
Each subfolder typically contains:
kwargs.jsonevaluate_all_k4.sh
Search-oriented examples may instead contain a Python driver plus a local README.md.
The example scripts run the released evaluator on all four k=4 benchmark metrics and print the resulting metrics.json.
Use these examples to:
- sanity-check the evaluator wiring
- compare known constructions on the same metrics
- inspect concrete canonical-potential parameterizations
Current examples include:
unifying_potential/unifying_potential_kplus1_competitiveness/unifying_potential_optimized_coefs/k3_circle_k4_generalization/search_n7_async_pipeline/: Codex-developed staged search for a fixedn=7index matrix that optimizes coefficient vectors and can reach a 3-violation candidate in roughly 20-30 minutes.