Pmtmax Autoresearch

Use this skill for the agent-driven autoresearch loop around lgbm_emos.

First reads

1. Read AGENTS.md.
2. Read:
   • docs/agent-skills/autoresearch.md
   • docs/agent-skills/research-loop.md
   • docs/codebase/modeling.md

Critical rules

• baseline is recency_neighbor_oof unless the run manifest says otherwise.
• create or edit one candidate YAML at a time under artifacts/autoresearch/<run_tag>/candidates/.
• use autoresearch-step for quick keep/discard/crash, then autoresearch-gate, then autoresearch-analyze-paper.
• canonical historical_training_set* / historical_backtest_panel stay immutable unless explicitly promoted elsewhere.
• public champion publish is never implicit; autoresearch-promote only promotes YAML specs and is the separate public-alias command.