Nature Social Science — Social Science Writing and Research Topic Justification Workbench
This skill transforms the actionable workflow for social science master's and doctoral research topic selection (derived from the ideas in the article Finding a PhD Topic with Codex) into a set of executable writing and self-check processes, covering quantitative, qualitative, ethnographic, archival, document-based, corpus, and digital research.
Scope of application: Social sciences and adjacent humanities disciplines. It applies to quantitative, qualitative, ethnographic, archival, document-based, corpus, and digital research.
The evaluation criteria are always the same: Is the question clear? Does the evidence match? Are the claims honest? Can the project be completed by this researcher?
What it does not do: It will not generate a "correct topic" for you, does not guarantee novelty, admission, funding, publication, or thesis success,
and does not treat any method as the default answer for high-quality research. AI confidence is not evidence; a completed form is not a pass certificate.
1. First, Routing: Which Stage Is the User At?
| User Request | Corresponding Element | Steps to Follow | References to Read |
|---|
| "I want to study XX" / "Help me find a thesis topic" | Topic Ideation | Step 1 → 2 → 3 | , |
| "Has anyone done this direction before" / How to synthesize literature / How to map the field | Literature Synthesis | Step 2 → 3 | |
| "I have data / I want to use method M" | Topic Ideation (false start) | Go back to Step 1 first | §6 Two false starts |
| "Can this topic be done" / Is there enough evidence | Argument Organization + Feasibility | Step 4 → 7 | , |
| "Help me generate several research questions" / How to write RQs | Topic Ideation | Step 5 | |
| "Help me think about how reviewers will critique this" / Check novelty | Argument Organization | Step 6 | |
| "Is the conclusion overstated" / Statement grading | Academic Expression | Step 4's claim ceiling + terminology discipline | 09-academic-expression.md
|
| "Help me revise / polish argument paragraphs" | Academic Expression | 对照 claim ceiling 与 Red flags | 09-academic-expression.md
|
| "How to talk to my advisor" / How to write opening materials | Delivery & Usage | Part III | |
| Need ready-made prompts | — | Take by step | (only authoritative version) |
| Need worksheet templates | — | Take by step | |
Entry judgment (mandatory first step): Only have an interest → Mode A; already have a broad theme or relationship → Mode B;
already have a tentative research question → Mode C (audit route, not advanced route).
If you only have a dataset or method preference, first ask "what unresolved phenomenon can it help understand?" If you can't answer, start with Mode A.
2. Four Iron Rules (must be followed in every response)
- Write "unknown" for things you are not sure about; do not let AI make up guesses.
- Actively seek contradictions, null results, inverse relationships, and alternative explanations — not materials that support your starting point.
- Retain evidence hierarchies, evidence statuses, failed searches, and key unknowns, do not smooth them into a single optimistic conclusion.
- Update after each step; do not carry entire old conversations into the next step.
Supporting prohibitions (to be included in every prompt provided):
- Do not generate topics, titles, or hypotheses (before the appropriate step);
- Do not claim the existence of research gaps or novelty;
- Do not default to quantitative, qualitative, causal, or any specific method;
- Do not write "no studies exist" when "not found in this round" is the case;
- Do not present personal interest as proven academic significance;
- Do not fabricate citations, sources, or results.
3. Twelve Judgments (Red Lines for Writing Self-Check)
Any judgment about topic selection or argumentation must first pass through this table. The full version is in §4 of
.
| Do Not Confuse | Actionable Check |
|---|
| Interest ≠ Topic | Have you defined the phenomena, processes, and boundaries to be explored? |
| Topic ≠ Research problem | Have you identified specifically where explanations fall short in the field? |
| Research problem ≠ Research question | Have you turned the knowledge difficulty into a task that can be advanced by evidence? |
| Research gap ≠ "few studies" | What knowledge consequences result from the scarcity of studies? |
| Novelty ≠ nobody has studied it | Does it resolve contradictions, mechanisms, measurement issues, or newly answerable questions? |
| Interesting ≠ Important | Beyond personal interest, whose understanding of what will the answer change? |
| Important ≠ Answerable | Is there appropriate, compliant, and time-matched evidence? |
| Available source ≠ Valid evidence | Do the fields, texts, or observations truly correspond to the target concepts? |
| Correlation ≠ Mechanism | Have you observed the process, or only seen phenomena occurring simultaneously? |
| Advanced method ≠ Strong contribution | Does the method solve the original knowledge difficulty? |
| Feasible study ≠ PhD-worthy project | Can it support a core puzzle and coherent development? |
| AI judgement ≠ Researcher judgement | Which decisions require the user's values, experience, and advisor's judgment? |
4. Eight-Step Workflow One-Page Overview
| Step | Main Output | Decision Question | Gate |
|---|
| 1. Research Search Space | One-page search space | Are concepts and boundaries sufficient for searching? | PASS / REVISE / STOP |
| 2. Research Landscape | Field landscape | Can you see multiple questions, explanations, and counterevidence? | Same as above |
| 3. Unresolved Problems | 3–7 problem cards | Is there consequential knowledge uncertainty? | Same as above |
| 4. Evidence Map | 1–3 evidence pathways | Can you observe the core question and maintain the claim ceiling? | Same as above |
| 5. Candidate RQs | 6–12 candidate cards | Are there substantially different, critique-able questions? | Same as above (PARK enabled) |
| 6. Academic Survival | Novelty Audit + Reviewer 2 | Do candidates survive academic critique? | Same as above |
| 7. PhD Feasibility | Feasibility dossier | Are there no unaddressed hard failures, and can it develop into a PhD project? | Same as above |
| 8. Final Shortlist | Up to three Topic Cards | Can you clearly explain trade-offs and the next verification step? | Same as above |
Step 5 briefly expands candidates and then narrows them down; this is not a purely linear pipeline.
Rollback rules: If the design can only provide weak proxies, wrong units, or inappropriate timing → return to Question;
if even the knowledge consequences of the question disappear → return to Problem.
5. Five Status Systems (Do Not Combine Into a Single Score)
| What Is Being Asked | Which System to Use | Possible Values |
|---|
| Can I proceed to the next step? | Workflow Gate | PASS / REVISE / STOP (+ PARK) |
| How far has this evidence pathway been verified? | Evidence State | NAMED / DOCUMENTED / ACCESS-CONFIRMED / TESTED |
| How does this candidate fare after academic critique? | Academic Survival | SURVIVE / REDESIGN / PARK / REJECT |
| What is the current status of this hard requirement? | PhD Feasibility Dependency | CONFIRMED / RESOLVABLE DEPENDENCY / UNKNOWN / FAIL |
| Can this candidate enter the final comparison? | PhD Feasibility Disposition | PROCEED TO COMPARISON / REDESIGN / PARK / REJECT |
Six Inequalities That Must Not Be Smoothed Over:
- — Unknown requires verification actions, not optimistic interpretation.
DOCUMENTED ≠ ACCESS-CONFIRMED
- — Passing a bounded critique does not prove that no similar research exists anywhere in the world.
- — What is missing is decisive evidence or conditions, not the idea itself.
PASS ≠ permanent approval
- — Rejection records preserve the closest existing work, failure conditions, and boundaries, preventing the same mistakes from being repeated later.
Do not create a "composite status". A candidate can perfectly well be:
Step 6 SURVIVE, evidence DOCUMENTED, access dependency UNKNOWN, Step 7 disposition PARK, and current step Gate REVISE.
These labels are not contradictory; they answer different questions. See
for details.
6. Work Rhythm (Repeat These Five Beats for Each Step)
- Read the current chapter and confirm the purpose and Gate of this step;
- Find the corresponding prompt in and prepare only the ;
- Write the output into the corresponding table in , do not treat long conversations as final records;
- Review results using quality checks and Red flags; stop when human judgment is required;
- Update , then proceed to the next step or exit temporarily according to .
Context control: Only carry
content when moving to each new segment, and actively discard
content.
specifies what to bring, what to discard, which judgments must be made by the user,
and what types of unknowns will block progress for each of the nine handoffs.
7. When You Must Stop and Find a Human
When you cannot judge the weight of a debate in your discipline, cannot obtain key full texts or data documentation, the research involves sensitive participants or high-risk materials,
method learning cost is unclear, or the project depends on a specific advisor, institution, or data provider — hand the current cards to the appropriate person.
AI can help you prepare questions, but cannot authorize on their behalf.
For typical scenarios and methods that require handing off to a human, see
and §5 "When to Find a Human" in
.
8. Usage Agreements
- Addressing in prompts: The prompts in are preserved verbatim from the original, and refer to themselves in the text as "I am using Codex".
Just copy them directly; if using in another agent, you can replace with the name of the current assistant, and leave everything else unchanged.
- The only authoritative version of the prompts is . The "short previews" in each step file are for illustration only; do not create a separate version for parallel maintenance.
- Quick / Deep tiers: Steps 2, 4, 6, and 7 offer Quick Run and Deep Run tiers.
Quick Run reduces first-round workload, but does not lower evidence, ethics, access, or hard-failure standards.
- Do not treat long conversations as records: All conclusions must be entered into the corresponding tables in and .
- Write "unknown" for anything uncertain, and do not accept optimistic completions from any party (including yourself).
9. Reference File Index
The following nine files are all located in the
directory of this skill. Read only the one needed for the current step as required.
| File | Content |
|---|
| Core concepts, six easily confused concepts, twelve judgments, capability boundaries, evidence hierarchy, three entry modes, Gate definitions, Funnel |
| Step 1 Search Space / Step 2 Field Landscape (core of literature synthesis) / Step 3 Unresolved Problems |
| Step 4 Evidence Map and claim ceiling / Step 5 Candidate Research Questions |
| Step 6 Academic Critique (Novelty Audit + Reviewer 2) / Step 7 Feasibility / Step 8 Final Shortlist |
| All copyable prompts (only authoritative version) |
| 11 worksheet templates |
| Five status systems + terminology glossary + Current Research State |
| Shortlist usage, advisor communication, nine handoffs, recovery rules |
09-academic-expression.md
| Academic expression discipline: claim ceiling, verb-evidence alignment, honest phrasing sentence pattern table, step-by-step Red flags (required reading for drafting and revising) |
Only do four things when using: Write "unknown" when you don't know; require searching for counterevidence; retain evidence hierarchies and failure records; update the status table at each step.