| name | autolab-post-run-reflector |
| description | Turn one directional-perps evaluation packet into a short evidence-linked decision memo with YAML frontmatter. Keep one next move, keep a short do-not-repeat list, and never write a diary-style run reflection. |
Post-Run Reflector
Goal: produce one compact decision memo for one evaluated candidate.
Rules:
- Tie every lesson to observed evidence in the packet.
- Keep a single
next_move.
- Keep
do_not_repeat short and concrete.
- Include the exact failed motif signature and ban repeating that motif when the result is non-informative or clearly bad.
- Keep the body extremely compact.
- State what changed versus the parent and why that change failed or held up.
- State one reusable lesson and one next test only.
- Do not write a run diary or a frontier summary.
- Use the
recent_completed_runs context only to avoid repeating motifs or to sharpen the next move.
- If the packet includes
return_driver, exposure_profile, or compact price/carry contribution fields, use them to explain what actually drove returns.
- For cross-sectional carry families, do not attribute causality to
trade_style unless the packet evidence explicitly supports that claim.
- Do not give generic advice. Name the exact motif, exact failed change, or exact successful change.
- Keep the body to four short lines:
What changed:Why it failed/worked:Do not repeat:Next test:
Output:
- use
templates/lesson_card.template.md
- frontmatter plus a short body only
