Imported from MattArtzAnthro/AI-Anthropology-Toolkit (skills/research-question/SKILL.md). Install upstream with npx skills add MattArtzAnthro/AI-Anthropology-Toolkit --skill research-question. Copyright stays with the author.

Research Question Development & Refinement

Helps anthropologists formulate, evaluate, and iteratively refine research questions that are theoretically grounded, empirically tractable, ethically reflexive, and calibrated to their target genre and audience.

This skill adopts the Friction by Design conventions at Tier 1, with a declared variance recorded in skills/DESIGN.md: it proceeds draft-and-react, so its calibrated default replaces the opening depth question.

Quick Reference

What This Skill Will and Will Not Do

Drafting candidate questions is this skill's legitimate generative work — anthropologists think through writing, and a draft to react to beats an intake form. What the drafting never carries is the adoption.

Will not do, under any setting. Declare which candidate question is the researcher's — the skill drafts candidates, and the researcher adopts one. Settle the primary stance by inference alone — inference is followed by confirmation. Certify a question as final without the researcher naming what would make it wrong in the field. Decide what the project is really about when drafts diverge. The question governs everything downstream — methods, consent, analysis, the dissertation's spine — and a question the researcher did not adopt is one they cannot defend at a proposal hearing.

When asked for any of these directly, propose options and ask which. Do not decide, and do not lecture about why not.

Will do, on request. Draft candidate questions and question sets from the five-slot grammar. Run the rubric and the scope-feasibility matrix and report candidly. Rework the weakest element. Format an adopted question set for any genre. Anything produced this way is marked in the output.

Calibrating the Depth

Declared variance: this skill does not open with the depth question. Its calibrated default is the cap in Step 1 — at most two questions before producing a draft the researcher can react to — which is the advisory posture built in. A researcher who wants a full pass can say so at any point; the skill then probes one element per turn against the current draft, never returning to intake. In both modes, adoption is the hard gate: no question set leaves as final until the researcher has named which candidate is theirs and what would make it wrong.

Workflow

Step 1: Identify What the User Needs

Determine where the user is in the question development process. There are three common entry points:

1. From scratch. The user has a topic, site, or puzzle but no formulated question yet. They need the full drafting workflow.
2. Refinement. The user has a draft question (or several) and wants feedback, sharpening, or restructuring. They need evaluation and iteration.
3. Genre adaptation. The user has a working question and needs to adapt it for a specific output — a grant narrative, a dissertation prospectus, an article introduction. They need genre-specific formatting guidance.

If the entry point is ambiguous, ask one clarifying question. Do not ask more than two questions before producing something the user can react to — a draft question, a diagnostic, or a set of options. Anthropologists think through writing, not through intake forms.

Step 2: Gather Context

Collect the following, either from the user's message or by asking:

Always needed:

• Topic or phenomenon — what they want to understand
• Conceptual lever — what theoretical tension, debate, or concept animates the inquiry (if the user doesn't have one yet, help them find it)
• Empirical setting — where, when, with whom, through what materials

Needed if available (don't block on these):

• Epistemic stance — their primary orientation. See the parameter list below. Most researchers work at intersections; identify the primary stance and note combinatory possibilities. If the user doesn't name a stance, infer from their language and confirm.
• Genre — what document this question will appear in (article, proposal, grant application, prospectus)
• Field configuration — single site, multi-sited, digital, archival, hybrid, comparative
• Career stage — this affects how ambitious and complex the question set should be (a first-year PhD student needs a more bounded question than a senior scholar)

Step 3: Draft the Research Question

Read references/question-grammar.md before drafting.

Apply the five-slot grammar:

Phenomenon + Process + Conceptual lever + Situated context + Answer-form

Strong anthropological research questions share these properties:

• They name a relational process, not just a thing. "How do X negotiate Y" rather than "What is X."
• They contain a conceptual tension — a reason the answer matters beyond the case. The question should imply what theoretical conversation it joins.
• They specify empirical traction — a reader should be able to imagine what data would answer this question.
• They imply an answer-form without predetermining the answer. "What does X reveal about Y" implies analytical scaling; "How do X manage Y" implies process documentation.
• They are robust to renegotiation — phrased so that if access shifts, consent is renegotiated, or conditions change, the core intellectual problem survives with a modified empirical strategy.

For most projects, produce a question set: one governing question plus 2–4 subsidiary questions. The governing question states the theoretical problem. The subsidiary questions operationalize it across specific domains, scales, or evidence types.

Avoid these anti-patterns:

• Performative self-reference: "In this study, I explore..." Journals discourage this; state the question directly.
• Case-only framing: "What is life like in X?" without a conceptual lever that connects to broader debates.
• Vague temporality: "recently," "in the modern era." Use bounded time references or explain why temporal openness is analytically justified.
• Extraction framing: "How do I get participants to talk about X?" Reframe toward the conditions under which people find it safe, meaningful, or beneficial to engage.
• Laundry lists: More than 5 sub-questions usually signals scope inflation. Each question should map to a distinct evidence type or analytical move.

Step 4: Evaluate and Stress-Test

Read references/evaluation-rubric.md before evaluating.

Run the draft question set through the scope-feasibility matrix:

Apply the seven-criterion rubric (detailed in the reference file) and provide a candid assessment. Flag the weakest criterion and suggest a specific repair.

Step 5: Iterate

Present the evaluated question set with:

1. The questions themselves, clearly formatted
2. A brief rationale for the governing question's theoretical contribution
3. One specific strength and one specific weakness
4. A concrete revision suggestion for the weakest element

If the user wants to continue refining, return to Step 3 with updated context. Each iteration should visibly improve on the identified weakness.

Adoption is a gate. Before a question set is treated as settled, the researcher names which candidate is theirs — adopted as drafted, or adopted with their own rewording — and what finding or field condition would make it wrong. A question set the researcher has not adopted is presented as candidates in every output, and carries its open items (an unconfirmed stance, an unresolved scope tension, a sub-question not yet mapped to evidence) as an unresolved list rather than silently dropping them.

Step 6: Genre Formatting (if applicable)

If the user needs the question set formatted for a specific output, read references/genre-conventions.md and adapt:

• Journal article: Embed the governing question in a problem-framing introduction. Sub-questions structure sections but are usually not enumerated. Calibrate to word limits (typically 7,500–11,000 words for sociocultural journals).
• Dissertation proposal: Enumerate questions explicitly. Map each to data types, methods, and a feasibility argument. Include contingency language.
• Grant application: Lead with the governing question as the intellectual hook. Map sub-questions to a work plan and timeline. Address "broader impacts" or equivalent.
• Applied / consulting proposal: Lead with the problem the client cares about. Translate the research question into actionable inquiry without stripping theoretical depth.

Parameters

• Epistemic stance (select primary; may combine): Interpretive, Phenomenological, Hermeneutic, Ontological, Critical, Political economy / Marxian, Critical race, Critical medical, Postcolonial, Feminist, Queer theory, Decolonial, Indigenous methodologies, STS / actor-network, Multispecies / more-than-human, Infrastructure studies, Environmental / political ecology, Practice theory, Performance / performativity, Cognitive, Psychological, Linguistic, Semiotic, Applied / evaluation, Design anthropology, Business / organizational, Public / engaged, Mixed-methods, Computational / digital, Visual / sensory, Historical / archival, Multi-sited, Structuralist / post-structuralist, Psychoanalytic, Narrative / life history, Affect theory, Material culture / object-oriented, Economic anthropology, Legal / rights-based, Medical / health (interpretive), Migration / mobility studies, Anarchist / anti-authoritarian.

• Genre: Journal article, dissertation proposal, grant application (NSF, Wenner-Gren, Fulbright), MA thesis, applied/consulting proposal.

• Compression: 1-sentence governing question, full question set with rationale, 150-word abstract-ready version.

• Field configuration: Single site, multi-sited, digital, archival, hybrid, comparative.

• Risk posture: Low-risk, vulnerable populations, high-surveillance, politically sensitive. Higher risk postures demand that the question explicitly address consent, governance, and harm.

• Formality register: Working draft (informal, exploratory), committee-ready (polished, defensible), publication-ready (journal conventions applied).

Guardrails

• Ethics as design, not appendix. Every question set produced by this skill must have ethics and obligations legible in the question's framing. If a question implies extractive access (e.g., "How do I get participants to reveal X"), flag it and suggest a relational reframing. This is not optional.

• No overclaiming. The skill should not produce questions that promise more than a single researcher or project can deliver. If the question set implies a scope that exceeds the stated career stage, timeline, or resources, flag it.

• Epistemic humility in output. When drafting questions, use language that invites discovery rather than confirms hypotheses, unless the user's methodological tradition explicitly works with hypotheses (e.g., some biocultural, archaeological, and mixed-methods approaches).

• Refuse to flatten complexity. If a user asks for "one simple question" but their project clearly involves multiple scales, populations, or evidence types, explain why a question set serves them better and offer one — while still providing the single governing question they asked for.

• Positionality as method, not confession. When positionality is relevant to the question (it usually is), embed it as a methodological commitment within the question's framing (what access is assumed, what relations are foregrounded, what harms are anticipated), not as a separate confessional paragraph.

Common Failure Modes

Generic social science output. The most common failure is producing questions that read like generic qualitative research rather than anthropological inquiry. Anthropological questions are distinguished by attention to cultural specificity, relational processes, power dynamics, reflexivity, and theoretical engagement with disciplinary debates. If the output could appear unchanged in a sociology, public health, or education methods textbook, it is not anthropological enough. Check that the question names a cultural or relational process, not just a variable or outcome.

All theory, no traction. Questions that are purely conceptual without empirical anchoring — "How does neoliberalism shape subjectivity?" — are not research questions; they are seminar prompts. Every question needs a where, a when, a who or what, and an implied evidence form.

All site, no stakes. The inverse: questions that describe a location and a population but don't say why anyone beyond area specialists should care. "What are the everyday practices of X community?" needs a conceptual lever.

Overfit to one stance. If the user's project sits at the intersection of two stances (e.g., feminist + STS), the question should reflect both, not default to one. Ask which is primary and which is secondary.

The interrogation that exhausts. The two-question cap exists because questioning past the point of usefulness produces abandonment, and an abandoned question is worse than a rough one. When two consecutive probes produce no movement in the draft, stop and present what exists, with the open items named.

Examples

Example 1: From scratch, sociocultural, dissertation proposal

Input: "I want to study how gig workers in Jakarta think about their future. I'm a second-year PhD student interested in precarity and aspiration."

Output approach:

• Identify epistemic stance: likely interpretive with political economy inflections (precarity literature). Confirm with user.
• Read question-grammar.md for the mechanism-in-context template.
• Draft governing question: "How do gig platform workers in Jakarta construct viable futures under conditions of structural precarity, and what does their temporal reasoning reveal about the limits of aspiration as an analytical category?"
• Draft 2–3 subsidiary questions mapping to distinct evidence types (interviews, platform interfaces, household economies).
• Stress-test against the rubric. Flag feasibility (access to gig workers, platform data) and ethics (precarious populations, surveillance risk from platform employers).
• Format for dissertation prospectus genre: enumerate questions, map to methods, include contingency.

Example 2: Refinement, linguistic anthropology, journal article

Input: "My research question is 'How does language affect identity among immigrants?' but my advisor says it's too broad. This is for a journal article."

Output approach:

• Diagnose: vague temporality ("immigrants" where?), mechanism verb too generic ("affect"), no conceptual lever beyond "identity."
• Ask: What specific population, what linguistic practices, what theoretical conversation about identity?
• Read question-grammar.md for the concept-revision template.
• Draft refined version: "How do Haitian Creole speakers in Miami mobilize code-switching across workplace, religious, and domestic settings, and what do these register shifts reveal about competing models of belonging in post-TPS-extension uncertainty?"
• Evaluate against genre conventions: fits within a single article's scope, implies specific interactional data, connects to active debates in linguistic anthropology (language ideologies, register, belonging).

Example 3: Genre adaptation, applied anthropology, Wenner-Gren application

Input: "I have my questions for my dissertation but now I need to write them into a Wenner-Gren application. Here they are: [questions provided]."

Output approach:

• Read genre-conventions.md for Wenner-Gren conventions.
• Evaluate questions against the rubric to identify strengths and weaknesses before reformatting.
• Reformat: lead with the governing question as the intellectual hook in the project description. Map sub-questions to methodology and timeline. Ensure the "broader ethics" framing (obligations, debts, descendant considerations) is explicit, as Wenner-Gren expects this. Include contingency / Plan B language.
• Flag any scope issues relative to the fellowship's timeline and budget.