Skip to content
Cload Cloud
Data & Analysis

recursive-research

Recursive research up to PhD level across any domain (science, tech, business, arts, humanities) with source tiering, WDM + Munger inversion for autonomous deci

View source More in Data & Analysis

What recursive-research Does

Recursive Research is an advanced research automation skill designed to conduct deep, multi-layered investigations across any academic or professional domain—from quantum physics to business strategy to art history. It leverages recursive query refinement and source credibility tiering to build comprehensive knowledge hierarchies that reach PhD-level depth. The skill is engineered for researchers, analysts, product strategists, and knowledge workers who need to move beyond surface-level information and build defensible, well-sourced arguments. By combining intelligent source evaluation with decision-making frameworks like Munger’s inversion and Wegmans’s Delphi Method (WDM), it enables users to autonomously explore complex domains and reach evidence-backed conclusions without manual literature review.

How to Install

Prerequisites

  • Python 3.8 or higher
  • Git installed on your system
  • API access to academic databases (optional but recommended): Google Scholar API, Semantic Scholar API, or arXiv API
  • Claude API key (for autonomous decision-making features)

Installation Steps

  1. Clone the repository

    git clone https://github.com/Anjos2/recursive-research.git
cd recursive-research

  2. Create a Python virtual environment

    python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

  3. Install dependencies

    pip install -r requirements.txt

  4. Configure API credentials

    • Create a .env file in the project root
    • Add your API keys: 
      CLAUDE_API_KEY=your_key_here
SEMANTIC_SCHOLAR_API_KEY=your_key_here
ARXIV_API_KEY=your_key_here

  5. Verify installation

    python -m recursive_research --version

  6. Run a test query

    python -m recursive_research query "machine learning bias in hiring" --depth=3

Use Cases

  • Due Diligence for Product Decisions: Product managers researching competitive landscapes or evaluating emerging technologies can use recursive research to automatically surface 50+ relevant sources across trade publications, academic papers, and market reports, organized by credibility tier
  • Thesis Development for Business Strategy: Strategy consultants building investment theses or market entry recommendations can leverage the Munger inversion framework to automatically stress-test assumptions and identify counterarguments across sources
  • Academic Literature Reviews: PhD students and researchers can automate the initial discovery phase of literature reviews, recursively exploring citation networks and identifying seminal papers without manually visiting 100+ websites
  • Patent Landscape Analysis: Intellectual property strategists can recursively map patent families, technical claims, and competitor innovations across international databases to identify white space and infringement risks
  • Innovation Forecasting: Innovation teams can use the recursive framework to explore emerging domains (e.g., synthetic biology, quantum computing) by automatically identifying key researchers, breakthrough publications, and institutional clusters

How It Works

Recursive Research operates through a four-stage architecture. First, query decomposition breaks down a user’s research question into constituent sub-questions using Claude’s reasoning capabilities. For example, “What is the state of AI in healthcare?” automatically decomposes into 8-12 focused queries like “What are the FDA approval pathways for AI diagnostics?” and “What are the current failure modes in clinical AI deployment?”

Second, multi-source search and retrieval executes these decomposed queries across heterogeneous sources—academic databases, preprint servers, industry reports, news archives, and regulatory filings. Each source returns ranked results that feed into the third stage. Third, source credibility tiering applies a weighted scoring system that evaluates source authority (peer review status, institutional affiliation, citation count), recency, domain specificity, and citation consensus. Sources are automatically classified into tiers: Tier-1 (peer-reviewed, high-citation academic sources), Tier-2 (industry analyses, well-established journalism), and Tier-3 (blog posts, social media, preliminary findings). The system maintains transparency about source uncertainty and flags contradictions between tiers.

Fourth, decision framework synthesis applies two complementary reasoning approaches. Munger’s inversion technique automatically flips research questions (“What would make this wrong?” instead of “What makes this right?”) to surface blindspots and biases in the collected evidence. Simultaneously, the Wegmans Delphi Method aggregates expert perspectives from the sources into probabilistic forecasts. The output is a hierarchical research report that shows not just findings, but the confidence bands around them and the logical dependencies between claims across sources.

Pros and Cons

Pros:

  • Reaches PhD-level depth autonomously without manual literature review or domain expertise
  • Source tiering ensures transparency about evidence quality and flags conflicting information
  • Recursive decomposition discovers insights a single search query would miss
  • Munger inversion built-in identifies blindspots and stress-tests conclusions
  • Works across any domain—science, business, humanities, arts—with consistent methodology
  • Outputs structured, exportable results compatible with Notion, Obsidian, Excel
  • Significantly faster than manual research: hours instead of weeks

Cons:

  • API costs scale with recursion depth; complex queries can cost $30-50+
  • Requires API access to academic databases for best results; some databases are paywalled
  • May over-index on readily available, English-language sources; misses research in other languages
  • Struggles with very recent breaking news (< 2 weeks old) or extremely niche topics with little published research
  • Output quality depends heavily on how well you decompose your initial question; vague questions produce vague results
  • Cannot conduct original primary research (surveys, interviews, experiments); works only with existing published sources
  • Tier-3 sources (blogs, preprints) can introduce noise if you’re not careful filtering results
  • Academic Literature Mapper: Automatically visualizes citation networks and identifies seminal papers and research clusters within a domain
  • Competitive Intelligence Gatherer: Applies similar recursive decomposition and source tiering specifically to market analysis, competitor tracking, and industry trends
  • Decision Framework Synthesizer: Implements structured decision-making tools (MECE, decision trees, scenario planning) on top of research findings
  • Citation Analysis Tool: Provides deep citation tracking and impact metrics to evaluate which research sources are most influential in a field
  • Evidence Evaluator: Specializes in assessing research quality, methodological rigor, and replicability of claims across sources

Alternatives

  • Manual Literature Review + Google Scholar: Free and flexible but labor-intensive; requires expert domain knowledge to decompose questions effectively; can miss important sources
  • ChatGPT/Claude Direct + Web Search: Fast for quick answers but lacks structured source credibility tiering and doesn’t recursively deepen research; prone to hallucination on niche topics
  • Subscription Research Platforms (Bloomberg Terminal, FactSet, LexisNexis): Provide high-quality curated sources and tiering but are expensive ($20K-100K+/year), require institutional access, and lack autonomous reasoning frameworks like Munger inversion
Glossary

Key terms

Source Tiering
A credibility classification system that ranks information sources (academic papers, reports, news articles) into hierarchical tiers based on peer review status, institutional affiliation, citation impact, and domain relevance. Tier-1 sources (peer-reviewed research) receive highest weight; Tier-3 sources (blogs, social media) receive lower weight but remain searchable for early signals.
Munger Inversion
A decision-making framework popularized by Charlie Munger that reverses the typical research question. Instead of 'What makes this true?', it asks 'What would make this wrong?' or 'What am I not seeing?' This technique systematically surfaces blindspots, biases, and missing evidence in an analysis. The recursive research skill applies this automatically to all findings.
Wegmans Delphi Method (WDM)
A structured forecasting technique that aggregates expert opinions from multiple sources into probabilistic estimates and consensus judgments. Rather than taking a single expert's view, it synthesizes perspectives across sources to estimate the likelihood of future outcomes or the credibility of competing claims. Used for quantifying uncertainty in research findings.
Query Decomposition
The automated process of breaking down a broad research question into smaller, more specific sub-questions. For example, 'What is the state of quantum computing?' decomposes into questions about hardware progress, algorithm development, practical applications, and corporate investment. This enables deeper, more structured exploration than a single monolithic search.
Recursion Depth
A parameter that controls how many levels of follow-up research the system performs. Depth-1 answers your initial question; Depth-2 explores sub-questions; Depth-3+ investigates implications and contradictions. Higher depth produces more comprehensive but slower, more expensive results. Most users find Depth-3 optimal for PhD-level insights.
FAQ

Frequently Asked Questions

What does 'recursive' mean in this context, and how is it different from a regular search?

Recursive research means the system breaks down your initial question into sub-questions, researches each one, and then automatically generates follow-up questions based on gaps or inconsistencies in the findings. Regular search just returns static results for a single query. Recursive approach mimics how expert researchers actually work—they read a source, notice a knowledge gap, formulate a new question, and dive deeper. This creates a dynamic exploration that reaches PhD-level depth without manual effort.

How do I know which sources are reliable? Does it automatically filter for bias?

The skill uses source tiering to assign credibility weights. Peer-reviewed academic papers score highest; well-researched industry reports score medium; blogs and unverified sources score lower. However, the system doesn't filter sources out—it makes their credibility transparent so you can see conflicting evidence and evaluate it yourself. You can weight results toward Tier-1 sources if you need maximum rigor, or include Tier-2/3 if you're exploring early-stage innovations. The system flags when sources contradict each other, which is often where the most interesting insights live.

Can I use this for domains outside academia—like business strategy or product decisions?

Yes. While it excels at academic domains, recursive research works equally well on business questions: competitive analysis, market sizing, technology trends, customer behavior patterns. The key is that your question must have substantive research available. Niche business topics (e.g., 'market size for AI in veterinary diagnostics') will return fewer results than broad topics, but the recursive decomposition often finds creative angles you wouldn't have thought to search manually.

What's the difference between WDM and Munger inversion? Which should I use?

Munger inversion is a critique tool—it deliberately looks for ways your conclusions could be wrong, surfacing blindspots. Use it when you need to stress-test a thesis or avoid groupthink. The Wegmans Delphi Method (WDM) aggregates expert consensus from sources into forecasts and probability estimates. Use it when you need a definitive answer or need to quantify uncertainty. Best practice: use both. Munger inversion finds what you're missing; WDM helps you decide what to believe.

How long does a recursive research query take, and how much does it cost?

Execution time depends on recursion depth (how many levels of follow-up questions to explore). A depth-2 query on a well-researched topic typically takes 2-5 minutes and costs $2-8 in API calls. A depth-5 query on a niche topic might take 15-30 minutes and cost $15-40. You can set recursion depth limits upfront to control cost and time. Most users find depth-3 offers the best return on investment for reaching PhD-level insights.

Can I integrate this with my existing research tools (Notion, Airtable, etc.)?

The skill outputs structured JSON and markdown by default, which is compatible with most tools. You can pipe results directly into Notion databases, import into Obsidian as markdown, or export to CSV for Excel/Airtable. There are community integrations for popular platforms. Check the GitHub repository's 'integrations' folder for maintained connectors.

What happens if I ask about a very new topic or niche domain where little research exists?

The system will transparently report 'low source density' and may elevate Tier-2 or Tier-3 sources (industry analyses, preprints, recent news) to fill gaps. This is honest about uncertainty rather than hallucinating fake sources. For truly novel domains, you often need domain expert input anyway; think of the skill as accelerating your expert conversations by pre-mapping the landscape.

Can I customize the source credibility tiers for my specific domain?

Yes. You can define custom tier definitions via configuration. For example, in biotech, you might weight preprints (bioRxiv) more heavily than in other fields because preprints are standard in biology. You can also blacklist or whitelist specific sources, institutions, or publication years. This flexibility is crucial for domain-specific research where general credibility rules don't apply.

More in Data & Analysis

All →
Data & Analysis

postgres

Execute safe read-only SQL queries against PostgreSQL databases with multi-connection support and defense-in-depth security.

sanjay3290
Data & Analysis

deep-research

Execute autonomous multi-step research using Gemini Deep Research Agent for market analysis, competitive landscaping, and literature reviews.

sanjay3290
Data & Analysis

CSV Data Summarizer

Automatically analyzes CSV files and generates comprehensive insights with visualizations without requiring user prompts.