Research Ideation

This is the idea-generation phase, the step before outlining. Outlining assumes the contribution is already decided. This skill is where the contribution gets decided. It turns a research area into a slate of candidate directions, ranks them, and hands the single surviving direction to outlining.

The deliverables are two files under .writing/:

1. .writing/ideation.md holds the candidate slate, the FINER scores, the cross-model adversarial pass result, and the rejected list.
2. .writing/ideation-brief.md holds the research-direction brief for the selected direction, formatted for outlining Step 1.

Iron rule: Generate the full candidate list of 15 to 20 directions before critiquing any of them. Premature convergence is the failure this skill exists to defeat. The first three ideas are never the best three.

Separation rule: Generation breadth and the accept verdict are strictly separated. The executor may drop a candidate only on an objective budget fact. Quality, novelty, and impact are decided by FINER scores plus a cross-model adversarial pass, never by the executor's solo taste.

Advisory rule: Every verdict here is advisory. Surface scores and objections to the user and let the user pick the surviving direction. Never auto-reject a candidate and never auto-mutate .writing/ state without the user's choice.

Never fabricate. Do not invent papers, arXiv IDs, DOIs, citations, or numbers to justify or to sink a candidate. Mark an unconfirmed reference [UNVERIFIED] and move on. Fabrication at ideation contaminates every downstream stage.

When To Use