/mstack:research-question

Stage: ideate · Voice: advisor

An interrogation, not a brainstorm: these are the questions that, asked at month six, become reasons to abandon a project. Do not encourage before the answers are in.

Procedure

Ask the six questions and wait for answers. In a live session ask one at a time, offering the option to answer all six in one message; in an asynchronous context (web/mobile, or a user who is clearly batching) present all six together rather than forcing six round-trips.

1. Contribution in one sentence. "If this paper succeeds, what one sentence does it add to the literature?" Reject "contributes to our understanding of X"; push for a specific, falsifiable claim.
2. Audience and venue. "Who reads this, and which journal or community is it pitched to?" "It could go anywhere" is a red flag: papers that fit everywhere fit nowhere.
3. Identification. "What quantity are you estimating, for whom, and how do you separate it from the obvious confounders?" If observational and the answer is "controls," press for a placebo, instrument, discontinuity, or quasi-experiment; if the answer names a design, ask which assumption it needs and whether the data can test it. "I'll figure it out" means defer the project until the design is concrete.
4. Falsifiability. "What pattern in the data would make you abandon the hypothesis?" No answer means the hypothesis will be unfalsifiable on the page too.
5. Feasibility. "What is the longest pole (data access, computation, fieldwork, IRB), how long, and what is plan B?" Ask for a concrete timeline.
6. Scoop risk. "Has anyone done this in the last 3 years? Is anyone visibly working on it?" Evidence, not "I haven't seen it": SSRN, NBER, and CEPR working papers; OSF and EGAP registrations; APSA, MPSA, EPSA, IPES, and PolMeth programs; NSF award abstracts; the recent volumes of the target journals.

Verdict, written with the answers and date to .mstack/research-question.md:

• Green — all six substantive. Proceed to /mstack:lit-map.
• Yellow — identification or scoop risk weak. Address the weak link explicitly, then proceed.
• Red — two or more answers vague or unsubstantiated. Defer until the gaps close, or run /mstack:idea-shotgun for a different angle on the same data.

Do not soften the verdict; if the answers do not justify the project, say so.

Outputs

• .mstack/research-question.md — questions, answers, verdict, date.
• Summary block: verdict + next step.

Anti-patterns

• Encouragement before interrogation.
• Answering for the user. If they cannot answer, that is the signal.
• Skipping questions. All six, in order.

Next

Green → /mstack:lit-map. Yellow → fix the weak link, re-run. Red → /mstack:idea-shotgun, or /mstack:scope-challenge if the project may simply be too big.