idea-shotgun

/mstack:idea-shotgun

Stage: ideate Voice: generative

When to invoke

You have data, a topic, or an unanswered question — and you're not yet committed to a specific angle. The shotgun produces variants so you pick from a set instead of from a hunch.

Procedure

1. Load context. Read any .mstack/research-question.md (if a prior attempt exists) and the data documentation if applicable.

2. Ask the user for the seed. One paragraph: what they have (data, theoretical interest), what they think the question might be, and what audience they imagine.

3. Generate 4–6 angles. Each must vary along one of:

   • Same data, different question.
   • Same question, different identification.
   • Same identification, different scope (population, time, treatment).
   • Same headline, different audience (IPE / comparative / methods / public).

4. Score each angle on a 4-cell card:

   ANGLE N: <one-sentence claim>
   --------
   Contribution:    <one sentence — what new sentence this adds to the literature>
   Identification:  <one sentence — how to separate the effect from the obvious confound>
   Cost:            <data-acquire weeks | analysis weeks | total months>
   Risk:            <highest single risk: data, identification, scoop, fit>
   

5. Rank. Score each on (contribution × tractability) / (cost × risk) qualitatively. Surface the top 2 to the user.

6. Save to .mstack/idea-shotgun-<YYYY-MM-DD>.md with all angles, including the ones not surfaced — they may be next year's paper.

Outputs

• .mstack/idea-shotgun-<date>.md — full angle set with cards.
• Summary block: top 2 angles with a one-line case for each, and a recommendation on which to take to /mstack:research-question.

Anti-patterns to refuse

• Variations on a theme. If five angles are all "the same question with one variable swapped," that's not divergence. Push for genuinely different framings.
• Defaulting to the user's seed. The shotgun is most valuable when at least one angle reframes the question entirely.