write-summary

name	write-summary
description	Stage 4 (final). Write a 1-paragraph natural-language summary of the regression results. Triggers when state.stage == "tabulated".

Write summary

Pre

Read state/state.json. Confirm state.stage == "tabulated" and state.artifacts.regression is populated.

Act

Ensure the output directory exists (mkdir -p summary), then generate a 1-paragraph summary at summary/summary.md. Format:

Over {n_obs} monthly observations, the portfolio earned a CAPM alpha of {alpha:.4f} ({t_alpha:.2f}) per month and a market beta of {beta:.4f} ({t_beta:.2f}). The R-squared was {r2:.3f}. Alpha [is / is not] statistically significant at the 5% level.

The "[is / is not]" decision must be made from the actual t-stat (|t| >= 1.96). Do NOT hedge. Do NOT add extra interpretation beyond what the numbers say.

Verify -- via subagent

Spawn the verifier subagent with the prompt: "Read summary/summary.md and state/state.json. Confirm every number in the summary matches state.artifacts.regression. Confirm the significance call matches the t-stat. Reply OK or list discrepancies."

If verifier returns anything other than OK: halt with the discrepancy as halt_reason.

Post

state.artifacts.summary_path = "summary/summary.md"
state.stage = "done"
state.last_action = {...}

Run python scripts/verify_state.py. We are done.