# Interpretive Orchestration: Epistemic Partnership for Qualitative Research

URL: https://research-memex.org/docs/toolkit/interpretive-orchestration-plugin
Description: Interpretive Orchestration gives qualitative research sessions a staged workflow, specialist agents, and enforceable checks.



*AI can automate coding, but interpretive depth demands deliberate partnership.*

Interpretive Orchestration is a Claude Code plugin for structured human-AI collaboration in qualitative research. It is not a tool for faster coding; it is infrastructure for deeper thinking.

<ToolMeta slug="interpretive-orchestration-plugin" runtime="Claude Code" note="Infrastructure for deeper thinking" />

Built at Imperial College London · Framework: [Cognitio Emergens](https://arxiv.org/abs/2505.03105) (Lin, 2025) — Partnership Agency for interpretive inquiry · Paper: [Interpretive Orchestration](https://doi.org/10.1177/14761270261448645) (Lin, 2026, *Strategic Organization*) — the methodology this plugin operationalizes · Starter project: [interpretive-orchestration-starter](https://github.com/linxule/interpretive-orchestration-starter) — a ready-to-use VS Code workspace with beginner-friendly settings.

<Warning>
  This plugin deliberately enforces methodological rigor. You cannot skip stages or bypass human judgment. If you want speed over depth, this is not the right tool. If you want to become a better interpretive researcher, welcome to the atelier.
</Warning>

## Why it matters [#why-it-matters]

Traditional AI approach:

* Give AI your data → Get instant analysis → Review output
* AI automates manual labor (the "calculator mindset")
* Faster but not necessarily deeper

Interpretive Orchestration approach:

* Build your foundation first → Collaborate with visible reasoning → Synthesize with tradition
* AI asks questions that deepen your reflexivity
* Scale and interpretive depth through Partnership Agency

> "Human and AI boundaries dissolve into unified epistemic system generating insights neither could produce independently."
> — Cognitio Emergens (Lin, 2025)

The philosophy: Friction over flow. If the plugin stops you, it's asking you to think.

## How it works [#how-it-works]

Both human and AI are apprentices to the craft tradition of interpretive inquiry. Neither masters the other. Both learn from craft principles: rigor, reflexivity, theoretical sensitivity, interpretive depth.

<Steps>
  <Step title="Solo practice" icon="palette">
    What you do: Manual coding of 10-15 documents, memo-writing, framework development

    AI role: `@stage1-listener` watches but doesn't intervene

    Why it matters: This stage builds the irreplaceable theoretical sensitivity that makes partnership meaningful. This stage is enforced: doors to the collaborative workspace stay locked until the stage is complete.
  </Step>

  <Step title="Side-by-side collaboration" icon="handshake">
    What you do: Guide AI, refine patterns, parallel theoretical + empirical streams

    AI role: `@dialogical-coder` with 4-stage visible reasoning:

    1. What I observe in the data
    2. What patterns I'm noticing
    3. What theoretical connections emerge
    4. What remains uncertain

    `@research-configurator` translates technical decisions to research language, so you do not need coding knowledge for this step.
  </Step>

  <Step title="Dialogue with tradition" icon="book-open">
    What you do: Interpret significance, synthesize findings

    AI role: `@scholarly-companion` asks tradition's questions (Gioia, Corley, Constructivist GT)

    Outcome: Work that can withstand methodological scrutiny
  </Step>
</Steps>

## Skills architecture [#skills-architecture]

Skills are auto-discoverable capability packages that Claude loads when relevant to your request. Twelve ship with the plugin. Six of them carry most of the traffic:

<KeyTable>
  | Skill                | Triggers                      | Purpose                                                                  |
  | -------------------- | ----------------------------- | ------------------------------------------------------------------------ |
  | `project-setup`      | "initialize", "new project"   | Socratic onboarding + config                                             |
  | `gioia-methodology`  | "data structure", "Gioia"     | Data structure building (1st→2nd→aggregate)                              |
  | `literature-sweep`   | "literature", "Stream A"      | Search + fetch + organize papers                                         |
  | `interview-ingest`   | "transcribe", "convert audio" | Audio/PDF conversion                                                     |
  | `deep-reasoning`     | "think through", "plan"       | [Sequential Thinking](/docs/toolkit/sequential-thinking-mcp) integration |
  | `paradox-navigation` | "tension", "both/and"         | [Lotus Wisdom](/docs/toolkit/lotus-wisdom-mcp) integration               |
</KeyTable>

The other six are `analysis-orchestration`, `coding-workflow`, `coherence-check`, `document-conversion`, `methodological-rules`, and `project-dashboard`. Full triggers are in `plugin/CLAUDE.md` in [the repository](https://github.com/linxule/interpretive-orchestration).

## Bundled MCPs [#bundled-mcps]

Two MCP servers ship with the plugin and start with it. Neither needs an API key.

| MCP                                                          | Purpose in Plugin                      | Command               |
| ------------------------------------------------------------ | -------------------------------------- | --------------------- |
| [Sequential Thinking](/docs/toolkit/sequential-thinking-mcp) | Structured reasoning chains            | `/qual-think-through` |
| [Lotus Wisdom](/docs/toolkit/lotus-wisdom-mcp)               | Navigate paradoxes, integrate tensions | `/qual-wisdom-check`  |

Markdownify was bundled until v0.2.1, when an SDK incompatibility took it out. Document and audio conversion now runs through MinerU or a manual pass instead.

Optional MCPs (require API keys):

| MCP                                | Purpose                          | Key Required      |
| ---------------------------------- | -------------------------------- | ----------------- |
| [MinerU](/docs/toolkit/mineru-mcp) | High-accuracy PDF parsing (90%+) | `MINERU_API_KEY`  |
| [Vox](/docs/toolkit/vox-mcp)       | Multi-model validation           | Provider API keys |
| [Exa](https://exa.ai/)             | Literature search                | `EXA_API_KEY`     |
| [Jina](https://jina.ai/reader/)    | Fetch articles                   | `JINA_API_KEY`    |
| Zotero                             | Bibliography management          | `ZOTERO_API_KEY`  |

The plugin's own dependency list still names Zen MCP for the multi-model row. Zen is retired here: [Vox](/docs/toolkit/vox-mcp) registers the same way and does the same job.

## Commands [#commands]

Seventeen commands, grouped the way the plugin groups them.

<KeyTable>
  | Command                       | Group    | Purpose                                                                            |
  | ----------------------------- | -------- | ---------------------------------------------------------------------------------- |
  | `/qual-init`                  | Project  | Initialize a project with Socratic philosophical onboarding                        |
  | `/qual-check-setup`           | Project  | Verify installation and MCP server status                                          |
  | `/qual-design`                | Project  | Configure research design for multi-case, longitudinal, or parallel-stream studies |
  | `/qual-status`                | Project  | Show the journey dashboard and progress through the stages                         |
  | `/qual-advance-stage`         | Project  | Advance to the next stage, with validation                                         |
  | `/qual-configure-analysis`    | Project  | Model selection, thinking budget, batch strategy, cost                             |
  | `/qual-stage1-guide`          | Stage 1  | Guidance for manual coding and building theoretical sensitivity                    |
  | `/qual-memo`                  | Stage 1  | Write analytical memos while insights are still forming                            |
  | `/qual-complete-stage1`       | Stage 1  | Validate the manual foundation and close the stage                                 |
  | `/qual-parallel-streams`      | Stage 2  | Run the theoretical and empirical streams                                          |
  | `/qual-characterize-patterns` | Stage 2  | Dimensional analysis: how patterns vary, and under what conditions                 |
  | `/qual-code-deductive`        | Stage 2  | Apply settled patterns to the remaining data                                       |
  | `/qual-synthesize`            | Stage 2  | Merge the two streams into unified patterns                                        |
  | `/qual-think-through`         | Analysis | Deep sequential reasoning                                                          |
  | `/qual-wisdom-check`          | Analysis | Navigate paradoxes, integrate contradictions                                       |
  | `/qual-examine-assumptions`   | Analysis | Surface the ontological and epistemological commitments underneath the work        |
  | `/qual-reflect`               | Analysis | Synthesis dialogue and reflexive pause                                             |
</KeyTable>

## Agents [#agents]

The plugin provides four specialized agents, each scoped to a particular stage of the atelier methodology.

<KeyTable>
  | Agent                    | Stage | Role                                                      |
  | ------------------------ | ----- | --------------------------------------------------------- |
  | `@stage1-listener`       | 1     | Thinking partner — asks questions, never suggests codes   |
  | `@dialogical-coder`      | 2     | 4-stage visible reasoning for reflexive coding            |
  | `@research-configurator` | 2     | "The Whisperer" — technical setup without jargon          |
  | `@scholarly-companion`   | 3     | Asks tradition's questions about theoretical significance |
</KeyTable>

`@research-configurator` ("The Whisperer") bridges research goals and technical implementation for non-coding researchers, covering model selection, cost estimation, batch processing strategy, saturation monitoring, and more. Invoke with `@research-configurator Help me set up my analysis`.

## Methodology hooks [#methodology-hooks]

The plugin enforces the sandwich methodology through automated hooks:

* PreStage2 - Blocks AI coding until Stage 1 foundation is complete
* PostFiveDocuments - Triggers interpretive pause every 5 documents
* EpistemicCoherence - Checks philosophical consistency

If you're blocked, it's asking you to think. That's intentional epistemic friction.

## Host support [#host-support]

<HostSupport
  rows="[
{ host: 'Claude Code (CLI)', status: 'full', note: &#x22;The plugin's native target. The starter project ships a VS Code workspace tuned for CLI use.&#x22; },
{ host: 'Claude Code (Desktop &#x22;Code&#x22; tab)', status: 'full', note: 'Same engine; identical commands and agents.' },
{ host: 'Codex CLI / Codex Desktop', status: 'adjacent', note: &#x22;The plugin runs inside Claude Code, but a Codex session can install it for you cross-host (`claude plugin marketplace add linxule/interpretive-orchestration && claude plugin install interpretive-orchestration@interpretive-orchestration`), after which you switch over to run the `/qual-*` commands and methodology hooks. The methodology itself — staged solo→partnership→tradition work, the Cognitio Emergens framework — is also portable on its own: Codex users can read the paper at https://doi.org/10.1177/14761270261448645 and recreate the workflow without the plugin's enforcement layer.&#x22; },
{ host: 'Claude Desktop (Cowork)', status: 'partial', note: 'A second marketplace entry, interpretive-orchestration-cowork, carries the same commands and skills into Cowork\'s sandbox. The two bundled MCP servers do not auto-load there; connect them by hand.' },
{ host: 'Cursor / other clients', status: 'none', note: 'Plugin format is CC-only.' },
]"
/>

The bundled MCPs ([Sequential Thinking](/docs/toolkit/sequential-thinking-mcp) and [Lotus Wisdom](/docs/toolkit/lotus-wisdom-mcp)) are reusable across any MCP client even when the plugin itself isn't.

## Install [#install]

New to Claude Code? `INSTALL.md` in [the repository](https://github.com/linxule/interpretive-orchestration) has complete setup instructions (Windows, macOS, Linux).

Want a pre-configured environment? Use the [Starter Project](https://github.com/linxule/interpretive-orchestration-starter): a ready-to-use VS Code workspace with beginner-friendly settings.

Already have Claude Code?

```bash
cd your-qualitative-project
claude
/plugin install linxule/interpretive-orchestration
/qual-check-setup  # Verify installation
/qual-init         # Begin Socratic onboarding
```

Then: Complete Stage 1 manual coding (10-15 documents) before beginning collaborative analysis with AI. This builds the irreplaceable theoretical sensitivity that makes partnership meaningful.

## Who should use this? [#who-should-use-this]

<Tabs>
  <Tab title="Ideal Users" icon="check">
    * PhD students learning qualitative methods
    * Researchers committed to methodological rigor
    * Those working with Gioia methodology or Constructivist GT
    * Anyone valuing depth over speed
    * Researchers who want defensible collaborative work with AI
    * Those interested in human-AI epistemic partnership
  </Tab>

  <Tab title="Not Ideal For" icon="xmark">
    * Quick content analysis needs
    * Quantitative researchers seeking automation
    * Those wanting fully automated coding
    * Projects prioritizing speed over rigor
    * Researchers uncomfortable with enforced reflection
  </Tab>
</Tabs>

## Part of Research Memex [#part-of-research-memex]

This plugin is the practical implementation of [interpretive orchestration](/docs/introduction/core-principles): the core philosophy that animates Research Memex.

The progression:

1. [Core Principles](/docs/introduction/core-principles) → Understand the philosophy
2. Individual MCPs → Learn the tools ([Sequential Thinking](/docs/toolkit/sequential-thinking-mcp), [Lotus Wisdom](/docs/toolkit/lotus-wisdom-mcp), [Vox](/docs/toolkit/vox-mcp))
3. Interpretive Orchestration Plugin → Apply it all to qualitative research

## Resources [#resources]

* GitHub: [linxule/interpretive-orchestration](https://github.com/linxule/interpretive-orchestration) - full documentation, including `INSTALL.md`, `QUICK-START.md`, `ARCHITECTURE.md`, and `DEPENDENCIES.md` under `plugin/`
* Framework Paper: [Cognitio Emergens](https://arxiv.org/abs/2505.03105) (Lin, 2025)
* Implementation Paper: [Interpretive Orchestration](https://doi.org/10.1177/14761270261448645) (Lin, 2026, *Strategic Organization*)

***

[← Back to the Toolkit](/docs/toolkit/carrel) · [Agentic workflows →](/docs/advanced-topics/agentic-workflows)