Scientific Brainstorming

---
name: scientific-brainstorming
description: "Frame a vague research direction and explore alternative questions or explanations before a method is fixed. Use for early brainstorming or a research pivot; use idea-creator for concrete contributions within an established problem."
---

<!-- EVO-INQUIRY:BEGIN -->
For research work, read [EvoInquiry](../shared-references/evo-inquiry.md),
or reuse it if already loaded and unchanged in this session. Apply relevant
principles within the requested scope. Ideation produces proposals, not verified
findings, novelty certification, experiment authorization or review acceptance.
<!-- EVO-INQUIRY:END -->

# Research problem framing and brainstorming

Turn a broad interest, puzzling observation or stalled approach into useful
research questions and the next investigation that could distinguish them.
An explicit request for a few ideas can end with a concise candidate table.

## Establish the question

Use the user's purpose, existing observations, constraints, relevant prior work
and resources. Separate real constraints from assumed or negotiable ones. Ask
only for missing information that would substantially change the investigation;
otherwise proceed with labeled assumptions. Do not invent an intended venue,
available GPU budget, dataset access or an obligation to produce positive results.

For an established problem, use [idea-creator](../idea-creator/SKILL.md) for
contribution candidates or [research-refine](../research-refine/SKILL.md) for a
method proposal. A full discovery workflow is a separate requested scope.

## Explore and check

Generate distinct problem framings or explanations before anchoring on a favored
solution when practical. Reuse known facts; this is not a ban on reading existing
evidence. Useful prompts include a simpler explanation, a measurement artifact,
a boundary condition, a reversed assumption or a structural analogy. Use only
lenses that change a prediction; renamed components are not distinct ideas.

For promising candidates identify:

- The question, why answering it matters, and what is already known.
- Key assumptions and competing explanations, including a credible simple case.
- Observations that would distinguish them, and what would weaken the proposal.
- A feasible next inquiry: targeted reading, data inspection, proof feasibility,
  simulation or a small study design, with its information value and resource needs.

Use [research-lit](../research-lit/SKILL.md) when the task needs literature
evidence. Check closest work and contradictory findings, then reopen exploration
if the evidence changes the framing. Label unsearched novelty as unknown; an
unfamiliar combination or a bounded search without a match is not proof of novelty.

Prioritize using relevance, discriminating power, checked differences, feasibility
and value even if the result is null. Keep disagreement and important unknowns.
Numerical scores are optional decision aids, not probabilities or proof of quality.
Do not force a single winner or a fixed number of ideas when evidence is insufficient.

## Deliver and hand off

Give the requested depth: candidate questions, reasons to retain or set aside each,
and the next decision-changing investigation. Separate observations, assumptions,
predictions and recommendations. Research can progress by disproving a mechanism,
finding a boundary, refining a question or stopping an uninformative route.

Persist only when the task needs an artifact or an existing workflow requires it.
Reuse the project's proposal or idea report and stable idea IDs. For full discovery,
hand off the problem-framing section to [idea-discovery](../idea-discovery/SKILL.md).
Use [experiment-plan](../experiment-plan/SKILL.md) after a proposal is sufficiently
fixed; brainstorming alone does not launch pilots or initialize project governance.