Add pricing experiment suite and cache measurement

[JaeLeex]

Summary

• Add hosted-agent pricing experiment tooling: pricing_loop, pricing_experiment_suite, cache_savings_experiment, and validate_combined_ledger.
• Improve funded_btcswp_combined_run.py with OpenRouter cache metrics, optional fill polling, and ledger validation hooks.
• Add modules/openrouter_usage.py for shared cache metric extraction.
• Add strategies/llm_monitoring.py for monitoring/hedge-heartbeat qualification runs.
• Fix hl trade --dry-run so it no longer requires a private key when --price is explicit.

Experiment commands

# Full local suite (skips live OpenRouter if no key)
python3 scripts/pricing_experiment_suite.py --suite all --skip-live

# Cache measurement (needs OPENROUTER_API_KEY)
NUNCHI_EXPERIMENT_ID=cache-$(date +%Y%m%d) python3 scripts/cache_savings_experiment.py --rounds 6

# Monitoring / hedge heartbeat qualification
python3 scripts/pricing_loop.py run --scenario pilot_monitoring --agents 1 --duration-seconds 300 --tick 10
python3 scripts/pricing_loop.py run --scenario pilot_hedge_heartbeat --agents 1 --duration-seconds 300 --tick 10

# Combined LLM + trade ledger join (live trades need maker/taker keys + --confirm)
python3 scripts/funded_btcswp_combined_run.py --price 24000 --model openai/gpt-4o-mini --maker-taker --confirm --validate-fills
python3 scripts/validate_combined_ledger.py --input-dir data/funded_combined_run
python3 scripts/pricing_aggregate.py --input-dir data/funded_combined_run

Validation

• python3 -m pytest tests/test_openrouter_usage.py tests/test_validate_combined_ledger.py tests/test_trade_command.py tests/test_cost_metering.py — pass
• python3 scripts/pricing_experiment_suite.py --suite combined_dry_run --skip-live — pass

Stack

• Base: Add hosted runtime metering upload sidecar #60 (hosted metering upload)
• Merge after: Unblock hosted agent cost E2E and MCP-safe trade #59 → Add hosted runtime metering upload sidecar #60 → this PR

Made with Cursor