| name | run-metrics |
| description | Compute numerical metrics for a simulation run (as opposed to the qualitative analysis done by analyze-run). Use when the user says "run metrics", "计算指标", or "compute metrics". |
run-metrics
Compute numerical metrics for a simulation run (as opposed to the qualitative analysis done by analyze-run).
Trigger
Use when the user asks to compute run metrics. See the trigger phrases in the frontmatter description.
Workflow
1. Determine data_dir
- Provided by the user → use it directly.
- Not provided → run
ls -t data/ | head -5 to list the most recent directories and let the user choose.
2. Run compute_metrics
python scripts/compute_metrics.py --data-dir school_03121031
The world name is inferred from the run config; no extra argument is needed.
3. Produce a summary
Read analysis/results/{data_dir}_metrics.json and output:
- Innate stats: distribution of income/deposit/relationship (min, max, mean, median).
- Per-year key metrics: average deposit change per year, fulfillment trend, social eval trend.
- Outliers: agents whose metrics are extremely high or low.
- Correlation insights: relationships between innate conditions and growth metrics (e.g. income vs deposit_diff).
Format the output as concise tables/lists so the user can grasp the overall picture quickly.
