科研 Skill

社科研究流程总控

alterlab-ssci-orchestrator

Coordinates a study through design, measurement, sampling, analysis, and inference gates with a shared Design Passport. Use for an end-to-end methodology review of a whole study.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

SSci Orchestrator — Route the Study, Hold the Passport, Enforce the Gates

Skill type: DISCIPLINE-ENFORCING (thin coordinator). This is the conductor of the social-science methods spine. It does no analysis. It walks a study through the gates in order, carries a single Design Passport between them, and refuses to let a stage run before its prerequisite gate has passed. For the actual routing/refusal logic, each gate owns its own skill; this skill owns sequence and hand-off.

The Core Rule

NO STAGE RUNS BEFORE ITS GATE PASSES. THE PASSPORT IS THE SINGLE SOURCE OF TRUTH.

A methods pipeline fails when stages run out of order — sampling sized before the design is fixed, a causal estimator run before the identifying assumption is pinned, a discussion written before the claim is audited. The orchestrator makes the order non-optional and makes every gate's verdict visible in one artifact.

When to Use This Skill

  • "Walk me through designing and analyzing this whole study."
  • "Run the methods pipeline / a full methodology review on my project."
  • "I have a research question, some data, and a survey — take it from design to inference."
  • The user's ask spans more than one gate/module and needs coordination.

Does NOT Trigger

The request is really about…Route toWhy not this skill
A single stage (design, measurement, sampling, or claim audit)the matching alterlab-ssci-*-gateTrigger the gate directly; no orchestration needed.
Running one analysis method (DiD, CFA, QCA, SNA, ABM, text)the matching analysis moduleThe module does it; the orchestrator only dispatches.
Multi-agent execution mechanics (fan-out, judge panels)alterlab-workflow-orchestrationGeneral composition engine; this skill is the social-science-specific spine on top of it.
A non-methods task (writing prose, literature search)alterlab-paper-writer / alterlab-deep-researchOutside the methods pipeline.

The Pipeline

research question
  └─> [design-gate]  ── pins design_type + identifying_assumption ──▶ (MANDATORY: BLOCK-capable)
        └─> [measurement-gate]   quantitative constructs ─┐
            [reflexivity-gate]   qualitative/interpretivist ─┴─▶ (measurement-stage analog)
              └─> [sampling-gate]  ── frame + method + size logic ──▶ (WARN default)
                    └─> [ANALYSIS MODULE]  ── dispatched by design_type / data (see routing table) ──▶
                          └─> [inference-gate]  ── audits claims vs design/sample ──▶ (MANDATORY: BLOCK-capable)
                                └─> reporting

At the measurement stage the orchestrator picks the paradigm-appropriate gate: alterlab-ssci-measurement-gate for quantitative constructs (reliability/validity/invariance), alterlab-ssci-reflexivity-gate for qualitative/interpretivist work (positionality + Lincoln & Guba trustworthiness). Mixed-methods runs both on the respective strands.

Analysis-module routing (by design_type + data in the Passport):

If the design / data is…Dispatch to
Complex-sample survey (weights / strata / clusters / FPC)alterlab-survey-analysis
Quasi-experimental / observational causal (DiD, IV, RDD, panel FE)alterlab-causal-inference
Clustered / nested / longitudinal (variance partitioning)alterlab-multilevel-models
Latent constructs, CFA / SEM / IRT / invariancealterlab-sem-psychometrics
Small-N configurational, set-theoretic (fsQCA/csQCA)alterlab-qca
Relational / network dataalterlab-sna
Simulation of interacting agentsalterlab-abm-mesa
Corpora / open-ended text (computational)alterlab-text-as-data
Qualitative coding of interviews / open text (human)alterlab-qualitative-analysis
Pooling effect sizes across studiesalterlab-meta-analysis
Interpretive / theory-buildingalterlab-qualitative-methods (existing)
Combined strandsalterlab-mixed-methods (existing)

Hand-off rule. When the request matches a row above, open your answer by naming that skill by its exact id (for example alterlab-survey-analysis) and saying in one line why it fits better, then stop applying this skill's procedure. Name it even when you also need to ask for missing input. If only part of the request is out of scope, name the skill for that part and do only the in-scope part. A brief general pointer is fine; the named skill does the work.

Cross-cutting: when the dataset has missing values, alterlab-missing-data runs before the analysis module (state the mechanism, multiply impute) so every downstream estimate pools correctly.

Full stage contract, PASS/WARN/BLOCK rules, and the analysis-module decision logic: references/pipeline_contract.md. A stdlib validator that reads a Design Passport and reports the next gate and whether it is fail-closed: scripts/passport.py.

The Design Passport

One YAML artifact, created at entry and appended by each gate. The orchestrator never overwrites a prior gate's fields; it only advances the stage and dispatches the next gate.

research_question: <text>
stage: <design | measurement | sampling | analysis | inference | done>
design_type: <set by design-gate>
identifying_assumption: <set by design-gate>
claim_type: <causal | associational | descriptive>
constructs: [<appended by measurement-gate — quantitative>]
positionality: <appended by reflexivity-gate — qualitative>
trustworthiness: {credibility, transferability, dependability, confirmability}  # reflexivity-gate
reflexivity_evidence: <appended by reflexivity-gate — qualitative>
sampling: {<appended by sampling-gate>}
analysis_module: [<chosen by the orchestrator from design_type / data>]
claims: [<audited by inference-gate>]
gate_log: [{gate: design, verdict: PASS|WARN|BLOCK, note: ...}, ...]

Gate verdict semantics

  • PASS — the gate's discipline is satisfied; advance.
  • WARN — an advisory issue; advance with the caveat recorded in gate_log and surfaced in the final report (default for measurement/sampling when non-fatal).
  • BLOCK — fail-closed; the pipeline cannot advance until resolved. Design-gate BLOCKs when no design/assumption can be named; inference-gate BLOCKs when a claim exceeds the design/sample.

Self-Check Before Advancing a Stage

  • Has the current stage's gate emitted a verdict recorded in gate_log?
  • If the verdict is BLOCK, has it been resolved (not bypassed) before advancing?
  • Is the analysis module chosen from design_type, not from the user's tool preference?
  • Does the Passport still carry every upstream field (nothing overwritten)?

References

  • references/pipeline_contract.md — stage-by-stage contract, verdict rules, analysis-module routing logic, worked run.
  • scripts/passport.py — stdlib Design Passport validator / next-gate reporter.

Part of the AlterLab Academic Skills suite.

示例提示词

帮我按阶段把这项社科研究从设计到报告的各道关卡依次走一遍。

所属插件

社会学、政治学与城市研究面向社会科学研究:问卷与访谈设计、质性编码、文本即数据、社会网络、ABM、比较政治与选举数据、RCT 与 DID、城市与区域分析。

安装该插件会一并获得同领域的 58 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

社科研究流程总控能做什么?

Coordinates a study through design, measurement, sampling, analysis, and inference gates with a shared Design Passport. Use for an end-to-end methodology review of a whole study。

如何安装社科研究流程总控?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「社科研究流程总控」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

社科研究流程总控需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

社科研究流程总控经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。