Research convergence

Research Convergence

Not just finding papers: converging on arguments. Deep research that builds toward a thesis through structured argumentation.

CRITICAL INTEGRITY RULE

NEVER fabricate claims or evidence chains. Every claim in the argument map must trace back to a real citation from an actual search. If evidence is thin, say so. Gaps are findings too.

Modes

Socratic Mode

Triggered by: "socratic mode", "help me refine my argument", "challenge my thesis", "play devil's advocate on my research"

Interactive back-and-forth dialogue to sharpen the user's research direction.

Workflow:

1. Ask the user to state their current research question or thesis
2. Probe assumptions: "What evidence supports this claim?", "What would falsify this?"
3. Identify hidden premises and unstated dependencies
4. Challenge weak links: "How do you account for [counter-evidence]?"
5. Suggest alternative framings or stronger formulations
6. Iterate until the argument structure is tight

Output:

• Refined thesis statement
• Argument structure (premise chain leading to conclusion)
• Identified assumptions that need empirical support
• Suggested next steps (searches, experiments, readings)

Full Research Mode

Triggered by: "full research", "deep dive", "comprehensive argument", "build the case for", "research convergence"

Comprehensive multi-round investigation that builds an argument map from the literature.

Workflow:

1. Round 1 (Broad scan): Search across all available connectors for the core research question. Identify major themes, key authors, seminal papers (15-30 papers).
2. Round 2 (Targeted deepening): For each major theme, conduct focused follow-up searches. Chase citations forward and backward on the most relevant papers (20-40 additional papers).
3. Round 3 (Counter-evidence hunt): Explicitly search for papers that contradict or complicate the emerging thesis. Search for failed replications, negative results, critical commentaries (10-20 papers).
4. Round 4 (Synthesis): Organize all findings into an argument map. Identify the strongest line of argumentation. Note where evidence is strong vs weak.

Output: argument-map.md (see Output Format below)

Flash Mode

Triggered by: "flash search", "quick scan", "flash mode", "5 minute scan", "quick overview"

Rapid preliminary scan for when the user needs a fast orientation.

Workflow:

1. Construct a single well-formed query from the research question
2. Search top 2-3 available connectors
3. Return 5-10 most relevant papers with one-sentence relevance annotations
4. Sketch a preliminary argument in 3-5 bullet points
5. Identify obvious gaps and suggest where to dig deeper

Output:

• Ranked paper list with relevance notes
• Preliminary argument sketch (3-5 claims with tentative evidence)
• Suggested next queries for deeper investigation

Argument Map Structure

The core output of Full Research Mode is argument-map.md:

# Argument Map: [Research Question]

## Thesis
[One-sentence refined thesis statement]

## Core Claims

### Claim 1: [Statement]
**Evidence for:**
- [Citation key]: [One-sentence summary of supporting evidence]
- [Citation key]: [One-sentence summary of supporting evidence]

**Evidence against:**
- [Citation key]: [One-sentence summary of counter-evidence]

**Strength:** Strong / Moderate / Weak
**Notes:** [Any caveats, conditions, or qualifications]

### Claim 2: [Statement]
...

## Synthesis
[2-3 paragraph narrative connecting the claims into a coherent argument]

## Gaps and Limitations
- [Gap 1]: No empirical evidence found for [specific sub-claim]
- [Gap 2]: Conflicting results between [citation] and [citation]

## Annotated Bibliography
[Citations organized by argument thread, not alphabetically]

### Supporting [Claim 1]
- **smith2024**: Smith et al. (2024). "Title." Journal. ([2-sentence annotation])

### Challenging [Claim 1]
- **jones2023**: Jones et al. (2023). "Title." Journal. ([2-sentence annotation])

Progressive Narrowing Strategy

Each search round builds on prior results:

• Round 1 queries: Broad terms from the research question
• Round 2 queries: Specific terms extracted from Round 1 abstracts, key author names, cited references
• Round 3 queries: Negation queries ("failure of X", "limitations of X", "critique of X"), replication studies, meta-analyses
• Round 4: No new searches, only synthesis of accumulated evidence

Track all queries and result counts for reproducibility.

Search Session State

Maintain session state across rounds:

• Queries executed and result counts per source
• Papers found, deduplicated, and retained
• Emerging themes and their paper clusters
• Current argument structure (evolves across rounds)

Integration Points

• literature-search: All search operations delegate to literature-search skill for actual query execution
• citation-management: Papers selected for the argument map are added to library.bib
• paper-drafting: The argument map directly informs introduction and discussion section structure
• peer-review: The argument map helps reviewers check whether claims are adequately supported

When to Recommend Each Mode

Situation	Mode
User unsure of direction, exploring ideas	Socratic
User has a clear question, needs full evidence base	Full Research
User needs quick orientation on a new topic	Flash
User wants to stress-test an existing argument	Socratic
User preparing literature review section	Full Research
User deciding whether a topic is worth pursuing	Flash