/deep-research

Handoff Example: deep-research → academic-paper

This example demonstrates how deep-research full mode, after completing research, hands off to academic-paper to begin paper writing.

Scenario Setup

The user has completed deep-research full mode on the topic "AI-Assisted Quality Assurance in Higher Education: A Comparative Analysis of Implementation Strategies in East Asian Universities." Below is a summary of the research outputs.

deep-research Output Summary

1. Research Question Brief (from research_question_agent)

### Primary Research Question
How do East Asian universities (Taiwan, Japan, South Korea) differ in their
implementation strategies for AI-assisted quality assurance, and what factors
explain the variation in adoption patterns?

### FINER Assessment
| Criterion   | Score | Justification |
|-------------|-------|---------------|
| Feasible    | 4/5   | Public data + policy documents available |
| Interesting | 5/5   | Timely: AI policy divergence across similar systems |
| Novel       | 4/5   | Few cross-national comparisons in this space |
| Ethical     | 5/5   | No human subjects; public policy analysis |
| Relevant    | 5/5   | Directly informs HEEACT and peer agencies |
| **Average** | **4.6/5** | |

### Sub-questions
1. What AI-QA tools and frameworks have been adopted by accreditation agencies
   in Taiwan, Japan, and South Korea?
2. What institutional and policy factors facilitate or hinder AI-QA adoption?
3. To what extent do implementation outcomes align with stated policy objectives?

2. Methodology Blueprint (from research_architect_agent)

- Paradigm: Pragmatist (comparative policy analysis)
- Method: Qualitative comparative analysis (QCA) + document analysis
- Data sources:
  - Policy documents (3 countries' QA frameworks)
  - Accreditation agency reports (HEEACT, JUAA/NIAD-QE, KCUE)
  - Published case studies and conference proceedings
- Analytical framework: Diffusion of Innovation (Rogers) +
  Institutional Isomorphism (DiMaggio & Powell)
- Validity: Triangulation across document types + member checking
  with QA practitioners

3. Annotated Bibliography (from bibliography_agent)

23 sources identified, 18 retained after quality screening:
- 6 peer-reviewed articles (Level II-III evidence)
- 4 accreditation agency reports (Level IV)
- 3 policy documents (Level IV)
- 3 conference proceedings (Level V)
- 2 institutional case studies (Level V)

Key themes identified:
- AI adoption maturity gap (Taiwan > Korea > Japan in QA applications)
- Regulatory divergence (Japan's cautious approach vs Korea's top-down push)
- Data infrastructure as prerequisite (not just technology)

4. Synthesis Report (from synthesis_agent)

## Core Findings

1. **Adoption patterns cluster into 3 models**:
   - Taiwan: bottom-up experimentation (agency-led pilots)
   - South Korea: top-down mandate (ministry-driven framework)
   - Japan: cautious incrementalism (voluntary guidelines)

2. **Critical success factors** (converging evidence across all 3 countries):
   - Data infrastructure maturity
   - Accreditation agency technical capacity
   - University willingness to share data
   - Clear governance framework for AI decisions

3. **Key tension**: efficiency gains vs transparency concerns
   - All 3 countries report resistance from faculty regarding "black box" decisions
   - Only Korea has attempted regulatory response (AI Ethics in Education Act 2025)

4. **Evidence gap**:
   - No rigorous outcome studies yet (all countries < 3 years into implementation)
   - Student/faculty experience data almost entirely absent

## Contradictions Identified
- Taiwan HEEACT reports "high adoption" but academic literature suggests
  most universities are still in pilot phase → definition of "adoption" varies

## Knowledge Gaps
- Impact on accreditation outcomes (pass/fail rates)
- Cost-effectiveness analysis
- Faculty perceptions and behavioral changes

5. Preregistration artifact handoff

This qualitative example has no completed preregistration artifact. The shell-capable dispatcher still runs the named deterministic builder and carries one explicit preregistration-artifact/1.0 receipt:

schema_version: preregistration-artifact/1.0
status: not_provided
artifact_id: preregistration-unavailable
relative_path: null
artifact_provenance: not_provided
source_artifact_sha256: null
source_artifact_size_bytes: null
source_content_sha256: null
source_content_utf8_bytes: null
declared_at: <explicit caller-held RFC3339 value>
record_digest: <builder-derived canonical digest>

The shell-capable dispatcher passes the schema-safe artifact ID and the canonical builder computes record_digest; the non-shell research architect never guesses a digest. There is no companion for this receipt. Academic-paper and every later pipeline handoff validate and carry the complete sidecar bytes unchanged. This explicit unavailable state means the manuscript/preregistration pair is not checked; it is not evidence of agreement.

Handoff Instruction

After completing deep-research, the user inputs:

Now help me write a paper based on this research

academic-paper intake_agent Auto-Detection

intake_agent Detection Process:

Scanning available materials...

[DETECTED] Research Question Brief
  - Source: deep-research / research_question_agent
  - Status: Complete (FINER 4.6/5)
  - Action: SKIP academic-paper Phase 1 (topic scoping)

[DETECTED] Methodology Blueprint
  - Source: deep-research / research_architect_agent
  - Status: Complete (pragmatist QCA + document analysis)
  - Action: SKIP academic-paper Phase 2 (methodology design)
  - Note: Will integrate into Method section directly

[DETECTED] Annotated Bibliography (18 sources, APA 7.0)
  - Source: deep-research / bibliography_agent
  - Status: Complete, quality-screened
  - Action: SKIP academic-paper literature search
  - Note: Will use as foundation for Literature Review section

[DETECTED] Synthesis Report
  - Source: deep-research / synthesis_agent
  - Status: Complete (3 core findings + 1 contradiction + 2 gaps)
  - Action: SKIP academic-paper analysis phase
  - Note: Will structure Findings and Discussion sections around synthesis

[DETECTED] Preregistration artifact sidecar
  - Source: deep-research dispatching layer
  - Status: not_provided (explicit unavailable receipt; no companion)
  - Action: VALIDATE and CARRY sidecar byte-for-byte
  - Advisory effect: manuscript/preregistration pair remains NOT CHECKED

Materials Assessment: COMPREHENSIVE
Recommended academic-paper mode: full (with accelerated pipeline)
Estimated sections already covered: 60-70%
Remaining work: Writing, formatting, argumentation, and polish

COMPREHENSIVE above describes the four prose research materials only. It does not certify preregistration coverage, cross-document agreement, or a clean manuscript.

Skipped Steps and Reasons

Post-Handoff academic-paper Actual Workflow

=== academic-paper: Accelerated Pipeline ===

Step 1: STRUCTURAL MAPPING
  [outline_agent]
  - Input: RQ Brief + Methodology Blueprint + Synthesis Report
  - Output: Complete paper outline, each section tagged with corresponding deep-research materials
  - Output example:

    I. Introduction
       - Context: AI in HE QA (from Synthesis background)
       - Problem: Cross-national variation unexplained
       - Purpose: Compare 3 East Asian models
       - RQ: [Directly cite RQ Brief]

    II. Literature Review
       - 2.1 AI in Quality Assurance (from Bibliography themes)
       - 2.2 Diffusion of Innovation framework (from Blueprint)
       - 2.3 Institutional Isomorphism (from Blueprint)
       - 2.4 East Asian HE systems comparison

    III. Methodology
       - 3.1 Research design: QCA + document analysis (from Blueprint)
       - 3.2 Case selection and data sources
       - 3.3 Analytical framework
       - 3.4 Validity and limitations

    IV. Findings
       - 4.1 Three adoption models (from Synthesis Finding 1)
       - 4.2 Critical success factors (from Synthesis Finding 2)
       - 4.3 Efficiency vs transparency tension (from Synthesis Finding 3)

    V. Discussion
       - 5.1 Theoretical implications
       - 5.2 Policy implications for accreditation agencies
       - 5.3 Practical recommendations
       - 5.4 Limitations (from Synthesis gaps + Blueprint validity)

    VI. Conclusion
       - Summary + Future research directions

Step 2: SECTION WRITING (Parallel)
  [review_writer_agent] → Literature Review
    - Based on Annotated Bibliography and Synthesis, transform into a fluent literature review
    - Not sequential summaries, but thematic integration

  [method_writer_agent] → Methodology
    - Expand the Blueprint into complete methodology paragraphs
    - Add technical details and procedural explanation for QCA

  [findings_writer_agent] → Findings
    - Expand the Synthesis's 3 core findings into evidence-supported arguments
    - Each finding accompanied by specific cases and data
    - Address contradictions flagged in the Synthesis

Step 3: DISCUSSION + BOOKENDS
  [discussion_writer_agent] → Discussion
    - Original discourse: dialogue between findings and theory
    - Does not repeat Findings; instead explains "why"
    - Integrates deep-research Knowledge Gaps as Future Research

  [bookend_agent] → Introduction + Conclusion
    - Introduction: from macro context to specific RQ
    - Conclusion: no new information, echoes RQ and looks to the future

Step 4: ABSTRACT + FORMAT
  [format_agent]
    - Abstract: 150-250 words, IMRAD structure
    - APA 7.0 full-text formatting
    - Reference list cleanup (based on Bibliography + newly added citations)

Step 5: SELF-REVIEW
  [review_agent]
    - Internal consistency check (RQ → Method → Findings → Discussion logic chain)
    - Citation completeness (every claim has a citation)
    - Format compliance
    - Verdict: Ready for submission / Needs revision

User-Facing Handoff Experience

academic-paper's Opening Message:

I've read your complete outputs from the deep-research stage, including:

• Research Question Brief (FINER 4.6/5)
• Methodology Blueprint (QCA + document analysis)
• Annotated Bibliography (18 quality-screened sources)
• Synthesis Report (3 core findings + 2 knowledge gaps)

Based on these materials, I've completed a preliminary plan for the paper outline. Here is the planned paper structure: [Display outline]

I have two questions that need your confirmation:

1. What is your target journal? (This affects word limits and formatting requirements)
2. Should the paper language be English or bilingual?

Once confirmed, I'll begin writing. Because the research foundation is already comprehensive, I can skip the early exploration stages and proceed directly to paper writing.

Notes

1. Not copy-paste: academic-paper does not directly copy deep-research outputs, but transforms them into the tone and format of an academic paper
2. May discover new issues: During the writing process, academic-paper agents may discover points missed by deep-research and will proactively supplement them
3. Still requires user confirmation: Target journal, language preference, specific formatting requirements still require user input
4. Review recommendation auto-connects: After paper completion, the user can continue with academic-paper-reviewer for formal review