🔬 ReproHub

Research Reproducibility Verification Platform

---

The reproducibility crisis is real. Only ~40% of psychology studies replicate successfully. Most researchers never check the numbers. ReproHub changes that — upload a paper and its dataset, and get a full reproducibility verdict in seconds.

---

📖 What is ReproHub?

ReproHub is an automated research reproducibility verification platform. It takes a research paper (PDF) and its underlying dataset (CSV), re-runs every statistical test from scratch, and tells you exactly how well the reported results hold up against the raw data.

No manual checking. No guesswork. Just evidence.

---

✨ Features

	Feature	Description
📄	Smart PDF Parsing	Upload any research paper — ReproHub extracts every statistical claim automatically
🤖	Regex-Based Extraction	Detects APA-style results: t-tests, ANOVA, correlations, regressions, and more
🔬	Real Statistical Re-runs	Actually executes the tests on your data — no simulation, no shortcuts
📊	Composite Scoring	Multi-factor verdict weighing p-values, effect sizes, and test statistics together
🗺️	Column Mapping UI	Fuzzy-matches paper prose to dataset columns; fully editable before verification
💡	Remediation Guidance	Pinpoints the weakest component when a claim fails — not just "failed"
📄	PDF Report Export	Download a professional reproducibility report for sharing or publishing
🔒	No API Keys Required	Fully open-source, runs locally, no external services needed

---

🚀 Quick Start

# Clone the repository
git clone https://github.com/Junaid-Ahmed-Rupok/ReproHub.git
cd ReproHub

# Install dependencies
pip install -r requirements.txt

# Launch the app
streamlit run app/main.py

👉 Live Demo: reprohub.streamlit.app

---

🔬 Supported Statistical Tests

Parametric Tests

Test	Effect Size Reported
Independent t-test	Cohen's d
Paired t-test	Cohen's d
One-way ANOVA	Eta-squared (η²)
Pearson Correlation	r
Linear Regression	R², Adj-R², per-coefficient p-values
Logistic Regression	McFadden pseudo-R², LR chi-square

Non-Parametric Tests

Test	Effect Size Reported
Mann-Whitney U	Rank-biserial correlation
Kruskal-Wallis H	Eta-squared approximation
Wilcoxon Signed-Rank	Rank-biserial correlation
Spearman Correlation	ρ (rho)
Chi-square	Cramér's V

---

📊 How Reproducibility is Scored

ReproHub uses a composite scoring model — not just p-value comparison.

Composite Score = (p-value agreement × 50%)
               + (effect size agreement × 30%)
               + (test statistic agreement × 20%)

Each component is scored 0–1 using exponential decay, so small differences are penalised gradually and large differences are penalised heavily.

Score	Status	Meaning
≥ 0.80	✅ Reproduced	Results align across all dimensions
≥ 0.55	⚠️ Marginal	Close but meaningful discrepancies exist
< 0.55	❌ Not Reproduced	Results do not hold up against the data
—	❓ Could Not Verify	Missing columns, unsupported test, or insufficient data

Why composite scoring? A claim with matching p-values but wildly different effect sizes (e.g. Cohen's d = 0.2 vs 0.8) should not be called "reproduced." The old p-value-only approach missed this. ReproHub doesn't.

---

🗺️ How It Works

 ┌─────────────┐     ┌──────────────┐     ┌─────────────────┐     ┌──────────────┐
 │  Upload PDF │────▶│ Extract      │────▶│ Map Columns     │────▶│ Re-run Tests │
 │  + CSV Data │     │ Claims       │     │ (Fuzzy Match +  │     │ (SciPy /     │
 └─────────────┘     │ (Regex/APA)  │     │  Manual Review) │     │  statsmodels)│
                     └──────────────┘     └─────────────────┘     └──────┬───────┘
                                                                          │
                     ┌──────────────┐     ┌──────────────────┐           │
                     │ Export PDF   │◀────│ Composite Score  │◀──────────┘
                     │ Report       │     │ + Explanation    │
                     └──────────────┘     └──────────────────┘

Step 1 — Upload: Provide a PDF paper and its CSV dataset.

Step 2 — Extract: ReproHub scans for APA-style statistical notation and pulls out every claim automatically.

Step 3 — Review: Check the auto-mapped column assignments, fix anything the fuzzy matcher got wrong, and confirm each claim.

Step 4 — Verify: ReproHub re-runs the actual statistical tests and scores each claim using composite scoring.

Step 5 — Report: Download a detailed reproducibility report with per-claim breakdowns, scores, and remediation advice.

---

📂 Project Structure

ReproHub/
├── app/                        # Streamlit web application
│   ├── pages/
│   │   ├── 1_upload.py         # File upload + claim extraction
│   │   ├── 2_review.py         # Column mapping + claim confirmation
│   │   ├── 3_dashboard.py      # Results visualisation
│   │   ├── 4_report.py         # PDF report generation
│   │   └── 5_about.py          # About page
│   ├── config.py               # App configuration
│   └── main.py                 # Entry point + navigation
│
├── core/                       # Core logic
│   ├── engine.py               # Statistical test engine (11 tests)
│   ├── extractor.py            # Regex-based claim extraction
│   ├── comparator.py           # Composite reproducibility scoring
│   ├── matcher.py              # Fuzzy column matching
│   ├── validator.py            # Claim validation
│   ├── remediation.py          # Remediation guidance
│   └── schema.py               # Shared data schemas
│
├── models/                     # Pydantic data models
│   ├── claim.py
│   ├── result.py               # Result model with composite scoring
│   ├── report.py
│   └── validation.py
│
├── utils/                      # Utility functions
│   ├── pdf_parser.py           # PDF text extraction
│   ├── fuzzy_matcher.py        # FuzzyWuzzy wrapper
│   ├── file_handlers.py
│   ├── report_generator.py     # PDF report generation
│   ├── visualizations.py
│   └── helpers.py
│
├── tests/                      # Unit tests
├── data/                       # Raw, processed, benchmark data
├── static/                     # CSS, images, templates
├── docs/                       # Documentation
├── requirements.txt
└── README.md

---

🛠️ Technology Stack

Layer	Technology
Framework	Streamlit
Language	Python 3.9+
Statistics	SciPy, statsmodels
ML / Encoding	scikit-learn
Visualization	Plotly, Matplotlib, Seaborn
PDF Processing	PyPDF, pdfplumber
Report Generation	ReportLab, Jinja2
Fuzzy Matching	FuzzyWuzzy + python-Levenshtein
Data Models	Pydantic v2
Deployment	Streamlit Cloud

---

📋 Requirements

pandas >= 2.0.0
numpy >= 1.24.0
scipy >= 1.10.0
statsmodels >= 0.14.0
scikit-learn >= 1.3.0
streamlit >= 1.29.0
plotly >= 5.17.0
matplotlib >= 3.7.0
pypdf >= 3.0.0
pdfplumber >= 0.10.0
reportlab >= 4.0.0
jinja2 >= 3.1.0
pydantic >= 2.0.0
fuzzywuzzy >= 0.18.0
python-Levenshtein >= 0.21.0

---

🤝 Contributing

Contributions are welcome. Here's how to get started:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/your-feature)
3. Make your changes with clear, documented code
4. Write or update tests where applicable
5. Commit with a descriptive message (git commit -m 'feat: add X')
6. Push to your branch (git push origin feature/your-feature)
7. Open a Pull Request

Please open an issue first for major changes so we can discuss the approach.

---

🗺️ Roadmap

• LLM-powered claim extraction (prose-level, not just APA notation)
• LLM-powered column mapping (semantic, not just fuzzy string matching)
• Claim deduplication (same result across abstract + results section)
• Support for Excel, SPSS, Stata, and .docx input formats
• Batch mode (verify multiple papers at once)
• Unit test coverage for all core modules

---

👨‍💻 About the Developer

Sarder Junaid Ahmed

Data Scientist & Machine Learning Engineer

Transforming complex data into strategic decisions through rigorous statistical modeling and production-ready machine learning systems.

Specializations: Statistical ML · Causal Inference · Trustworthy AI · Fairness-Aware ML · RAG Systems

Selected Research:

• 📄 Ahmed, S.J. et al. (2026). Machine Learning for Crime Classification: A Fairness-Aware Approach to Class Imbalance. Journal of Machine Learning and Applications, 2(1), 9–17. DOI: 10.61577/jmla.2026.100002
• 📄 Ahmed, S.J. et al. (2026). CF-EGAT: A Causal Fairness-Aware Equity Graph Attention Network for Country-Level Environmental Livability Classification. SPECTRA 2026. 🏆 1st Best Paper Award
• 📄 Ahmed, S.J. (2025). Multi-Dimensional Statistical Similarity for Governance Classification: Beyond Arbitrary Thresholds. APMEE 2025. 🏆 Best Research Paper Award

Other Deployed Projects:

• 🔬 ReproHub — Automated research reproducibility platform with composite scoring across 11 statistical tests
• 📊 StatsPro — AI-powered statistical analysis platform with automated CSV-to-report workflows

Honors: 🏆 1st Best Paper — SPECTRA 2026  ·  🏆 Best Research Paper — APMEE 2025  ·  🎖️ Esteemed Alumni Award — YLRL RUET 2024  ·  ⭐ Perfect GPA 5.00/5.00 — SSC & HSC  ·  🎓 National Merit Scholarship — 2009 & 2013

---

📝 License

This project is licensed under the MIT License — see LICENSE for details.

---

🙏 Acknowledgments

Built in response to the reproducibility crisis in scientific research. Powered entirely by open-source Python libraries.

Built with Streamlit · LangChain · Groq · FAISS · sentence-transformers

---

Made with ❤️ for open science and reproducible research