Autoresearch: expansion strategy optimization

[realityinspector]

Summary

Infrastructure for Karpathy-style autoresearch applied to Clockchain expansion strategy optimization. The autoresearch loop mutates expansion parameters and measures resulting graph quality to find optimal configurations autonomously.

What's included (branch: autoresearch/clockchain/expansion)

• autoresearch/clockchain_autoresearch.py — main autoresearch loop with:
  • Mutation space: prompt templates, edge confidence thresholds, max edges per expansion, geographic/temporal/causal depth weights, edge type ratios
  • Simulated annealing temperature schedule
  • Dry-run mode with synthetic graph generation
  • JSONL results output + Pareto front JSON
• autoresearch/graph_metrics.py — graph quality metrics:
  • Edge density
  • Average causal chain depth
  • Clustering coefficient
  • Temporal coverage (span of timestamps)
  • Geographic diversity (unique locations)
• Composite scoring function balancing all metrics

Next steps

• Implement live mode: connect to Clockchain API to run real expansion cycles
• Integrate with actual expansion prompt templates in the codebase
• Add SNAG-Bench GCQ axis as an additional scoring signal
• Run overnight optimization sessions and analyze Pareto frontiers
• Apply winning configs to production expansion strategy