Jeuners
|
94025e03a0
|
Add analyze.py: time-dilation, tool usage, decision modes, cost
Reads emergence.db and prints a report covering:
- Time Dilation: tau, pace, gamma, |dtau| per agent and pairwise
(divergent markers when |dtau| > 3.0)
- Tool Usage: distribution with bar chart and percentages
- Decision Modes: llm vs fallback split, per-agent LLM-rate
- Latencies: per-model avg, p50, p95, min, max
- Generated Texts: recent blogs + billboards
- Constitution: current version + accepted amendments
- Cost Estimate: OpenRouter tokens projected to 24h
Usage:
python3 analyze.py
python3 analyze.py --db /path/to/other.db
python3 analyze.py --json
Live-verified report on the running system:
- 4 of 6 agent pairs DIVERGENT (gamma=0.82-1.12, |dtau|=5-16)
- llama3.2:3b latency: avg=7.11s, p95=14.37s (LAN @ 192.168.1.245)
- claude-haiku latency: avg=3.58s (OpenRouter)
- gpt-4o-mini latency: avg=1.33s (OpenRouter, fastest)
- Tool mix: 58% go_to_place, 22% write_blog, 20% memory
- Per-agent LLM rate: lovely/spark 100% (Ollama), anchor/flora
15% (OpenRouter rate-limited)
|
2026-06-15 11:33:53 +02:00 |
|