# Lumina Reviewer — Complete Documentation

> Comprehensive documentation for Lumina Reviewer: AI-assisted systematic literature review and citation screening software. Includes architecture, active learning mechanics, database integrations, dual-review workflows, PRISMA 2020 tracking, pricing, and database export tutorials.

---

## 1. Executive Overview

Lumina Reviewer (https://luminareviewer.com) is a modern web application for systematic literature reviews, scoping reviews, rapid reviews, and evidence syntheses. Developed for biomedical researchers, epidemiologists, systematic review authors, and academic research teams, it solves the primary bottlenecks in evidence synthesis:
1. **Screening volume**: Research teams often face thousands of candidate citations. Lumina continuously prioritizes the screening queue using semantic text embeddings and active learning classification.
2. **Cold-start delay**: Some classifier-based approaches need initial labeled examples before they can train; the number depends on the implementation and dataset. Lumina uses dense vector representations (`text-embedding-3-small`) to rank citations relative to review criteria immediately from paper #1.
3. **Multi-database evidence gathering**: Lumina integrates direct search across 7 scholarly and grey literature databases with deduplication and PRISMA 2020 tracking.
4. **Dual screening & auditability**: Reviewer-blind human screening, advisory Pixel-Bot recommendations, conflict adjudication, and PRISMA 2020 flow diagrams.

---

## 2. Core Features & Technical Architecture

### 2.1 Active Learning & Vector Ranking Engine
- **Embedding Pipeline**: Uses OpenAI `text-embedding-3-small` (1536-dimensional embeddings) to convert project criteria, paper titles, and abstracts into dense vector space.
- **Queue Prioritization**: Cosine similarity against review eligibility criteria provides immediate ranking for unreviewed papers from paper #1.
- **Online Classifier Update**: As reviewers mark studies as Include or Exclude, a fast probabilistic classifier (TF-IDF + Naive Bayes) retrains dynamically to learn project-specific inclusion criteria.
- **Transparency & Control**: Reviewers retain 100% manual decision authority. AI scores and confidence rankings serve strictly as advisory sorting aids.

### 2.2 Pixel-Bot AI Co-Reviewer
- Provides advisory title/abstract recommendations. It does not replace a second independent human reviewer or make final eligibility decisions.
- Evaluates title and abstract against user-specified inclusion and exclusion criteria.
- Outputs structured recommendations (Include, Exclude, Maybe) accompanied by explicit reasoning and extracted rationale quotes.
- Ideal for pilot testing screening criteria, solo researchers conducting exploratory reviews, and training novice screeners.

### 2.3 Unified Multi-Database Search
Lumina enables searching across 7 major scientific and grey literature repositories simultaneously:
1. **PubMed / MEDLINE**: Biomedical and life sciences literature via NCBI E-Utilities.
2. **OpenAlex**: Global scholarly graph with over 250 million scholarly records.
3. **Europe PMC**: Life sciences literature, full-text open-access articles, and clinical guidelines.
4. **arXiv**: Quantitative biology, statistics, computer science, and physics preprints.
5. **bioRxiv / medRxiv**: Biology, medicine, health science, and clinical preprints.
6. **OpenAIRE**: European research information system and open-access repositories.
7. **Government & Grey Literature**: Official publications and reports from regulatory bodies and public health institutes.

### 2.4 Deduplication Algorithm
Citation import compares normalized DOIs first, then normalized titles using SequenceMatcher similarity (threshold 0.85), then an author/year/partial-title fallback. Conflicting nonempty DOIs prevent title-only merging. Guards separate trial registrations and news from research articles. Matching is heuristic and can make mistakes; retain source exports and check removed records and counts. The public citation import benchmark demonstrates a controlled overlap example, not real-world duplicate-detection accuracy.

### 2.5 PRISMA 2020 Interactive Generator & Flow Tracking
- Real-time PRISMA 2020 flow diagram that tracks identification (records from each database), deduplication, screening exclusions, full-text eligibility assessments, and final inclusions.
- Standalone free generator at `https://luminareviewer.com/tools/prisma-flow-diagram-generator/` produces publication-ready diagrams with vector SVG and high-resolution PNG export.

### 2.6 Open-Access Full-Text PDF Retrieval
- Integrated with Unpaywall API and DOI resolvers to discover open-access PDFs automatically.
- Built-in split-screen PDF viewer within the screening interface allows reviewers to inspect full texts without leaving the queue.

### 2.7 Early Stopping Strategies
- Built-in stopping heuristics based on screening history and Work Saved over Sampling (WSS). They do not establish recall against all relevant studies.
- Reviewers can track the rate of new relevant papers found per block of 50 screened citations to inspect diminishing returns. Low yield does not prove that unscreened records are irrelevant; any stopping decision must follow the protocol.

---

## 3. Pricing & Subscription Plans

Lumina Reviewer offers transparent, researcher-friendly pricing without long-term contracts or institutional lock-in:

### Free Tier (€0)
- Interactive screening demo with PTSD benchmark dataset.
- Free online PRISMA 2020 Flow Diagram Generator (SVG/PNG downloads).
- Preview workspace with empty project creation and citation file import checks.

### Starter Plan (€9 / month)
- Designed for: Solo researchers, students, and thesis authors.
- 1 active project at a time.
- Up to 750 papers per project.
- Active learning queue prioritization from paper #1.
- Multi-database deduplication.
- Full export to CSV and RIS formats.

### Duo Plan (€19 / month)
- Designed for: Larger reviews and ongoing research, whether working alone or with a colleague.
- Two reviewer accounts included in one shared workspace; one person pays.
- Your colleague needs no separate paid subscription.
- Independent dual screening and conflict resolution.
- Project capacity is shared through the owner’s subscription.
- Up to 100 active projects.
- Up to 25,000 papers per project.
- Automated open-access full-text PDF retrieval via Unpaywall.
- Priority customer and methodology support.
- Advanced export formats.

### Team Plan (€20 / month base, includes 1 seat; +€7/month per additional seat)
- Designed for: Multi-reviewer research labs, systematic review teams, and clinical guidelines panels.
- Up to 100 active team projects.
- Up to 100,000 papers per project.
- Shared collaborative projects.
- Blind dual screening (reviewers cannot see each other's votes until unblinded).
- Disagreement flagging and conflict resolution adjudication dashboard.
- Full PRISMA 2020 team audit trail logging reviewer votes, timestamps, and consensus notes.

---

## 4. Step-by-Step Export Tutorials for Scholarly Databases

When importing search results into Lumina Reviewer, select the correct export format and ensure abstracts are included:

### 4.1 Export from PubMed
- **Recommended format**: Native `.nbib` citation-manager export for screening.
- **Workflow**: Run query → Send to → Citation manager → Confirm selected records or all results within the export limit → Create file.
- **Abstract check**: Inspect AB fields where the source supplies an abstract. CSV is a separate spreadsheet export and does not preserve the same abstract fields.

### 4.2 Export from Scopus (Elsevier)
- **Recommended format**: RIS format.
- **Export limit**: Up to 20,000 records per export.
- **Workflow**: Run search → Select records → Click "Export" → Select "RIS Format" → Under "What information do you want to export?", check "Citation information" AND "Abstract & keywords" → Click "Export".

### 4.3 Export from Web of Science (Clarivate)
- **Recommended format**: RIS format (`.ris` or `.txt`).
- **Export limit**: 1,000 records per batch. For larger result sets, export in intervals (e.g. 1-1000, 1001-2000).
- **Workflow**: Run query → Click "Export" → Select "Other File Formats" → Set record range → Set Record Content to "Full Record" → Set File Format to "Other Reference Software (e.g. RIS)" → Click "Export".

### 4.4 Export from Embase (Elsevier)
- **Recommended format**: RIS format.
- **Export limit**: 500 records (unauthenticated) up to 10,000 records (institutional login).
- **Workflow**: Run query → Check "Select all" → Click "Export" → Choose "RIS format (Mendeley, EndNote)" → Select "Full record" (or Citation + Abstract + Index Terms) → Click "Export".

### 4.5 Export from CINAHL (EBSCOhost)
- **Recommended format**: Direct Export in RIS Format.
- **Workflow**: Run search → Click "Share" → Add results to Folder (or use "Export results" for automated email delivery up to 25,000 records) → Open Folder → Select all → Click "Export" → Choose "Direct Export in RIS Format" → Ensure "Detailed citation and abstract" is enabled → Click "Save".

### 4.6 Export from Cochrane Library (Wiley)
- **Recommended format**: PC (RIS) or Mac (RIS).
- **Export limit**: Up to 20,000 trial records from CENTRAL.
- **Workflow**: Run search → Select either Cochrane Reviews or Trials (CENTRAL) tab → Check "Select all" → Click "Export Selected Citations" → Choose "PC (RIS)" → Check the critical "Include abstract" box → Click "Export".

### 4.7 Export from Google Scholar
- **Methods**:
  - *Method A (My Library)*: Star results → Open "My Library" → Select all → Click "Export all" → Choose "RefMan" (`.ris`).
  - *Method B (Publish or Perish)*: Query Google Scholar through Harzing's desktop software → Click "Save results" → Choose "RefMan/RIS format".

### 4.8 Export from PsycINFO (APA PsycNet / EBSCOhost / Ovid)
- **Recommended format**: RIS format.
- **Workflow**:
  - *APA PsycNet*: Select results → Click "Export" → Format: "RIS" → Content: "Citation and Abstract" (or "Complete Reference") → Click "Continue".
  - *EBSCOhost*: Add to folder → Export → "Direct Export in RIS Format" with "Detailed citation and abstract".
  - *Ovid*: Select range → Click "Export" → Format: "RIS" → Content: "Complete Reference" → Click "Export".

---

## 5. Product Capabilities and Comparison Scope

| Lumina capability | What it does | Limit |
| :--- | :--- | :--- |
| Active learning | Prioritizes records against review criteria | Human reviewers decide eligibility |
| Pixel-Bot | Advisory Include, Exclude or Maybe recommendations | Does not count as an independent human reviewer |
| Duo and Team | Independent human screening and conflict resolution | Requires two human assessments for dual screening |
| Citation import | Normalizes titles, abstracts, authors, dates, journal, DOI and source ID | Does not recreate external reviewer decisions, notes or audit logs |
| PRISMA tools | Tracks review counts and generates flow diagrams | Reporting does not establish methodological compliance |

For comparisons with Rayyan, Covidence, ASReview and DistillerSR, see the public comparison pages and check each vendor’s current documentation and pricing. Competitor capabilities and prices change; this file does not assert an independently verified competitor ranking.

---

## 6. Complete Resource & Guide Index

### Free Tools
- PRISMA Flow Diagram Generator: `https://luminareviewer.com/tools/prisma-flow-diagram-generator/`
- Multi-Database Academic Search: `https://luminareviewer.com/tools/multi-database-search/`
- Interactive Demo: `https://luminareviewer.com/demo/`

### Methodology Guides
- Systematic Review Protocol: `https://luminareviewer.com/guide/systematic-review-protocol/`
- Complete Guide to Systematic Reviews: `https://luminareviewer.com/guide/how-to-do-a-systematic-review/`
- PICO Framework Explained: `https://luminareviewer.com/guide/pico-framework/`
- Eligibility Criteria Design: `https://luminareviewer.com/guide/inclusion-exclusion-criteria/`
- Title & Abstract Screening Best Practices: `https://luminareviewer.com/guide/title-abstract-screening/`
- Active Learning in Evidence Synthesis: `https://luminareviewer.com/guide/active-learning-systematic-reviews/`
- Dual Screening & Conflict Resolution: `https://luminareviewer.com/guide/dual-screening/`
- PRISMA 2020 Reporting Checklist: `https://luminareviewer.com/guide/prisma-2020-checklist/`
- PRISMA Flow Diagram Step-by-Step: `https://luminareviewer.com/guide/prisma-flow-diagram-explained/`
- Early Stopping Strategies & Target Recall: `https://luminareviewer.com/guide/stopping-strategies/`
- Full-Text PDF Management: `https://luminareviewer.com/guide/full-text-pdf-retrieval/`
- Grey Literature Search Strategies: `https://luminareviewer.com/guide/grey-literature-search/`
- Search Results Deduplication: `https://luminareviewer.com/guide/deduplicate-systematic-review-search-results/`
- RIS Citation File Guide: `https://luminareviewer.com/guide/how-to-import-ris-files/`
- Scientific Database Ingestion: `https://luminareviewer.com/guide/import-from-scientific-databases/`

### Database Export Guides
- PubMed Export Guide: `https://luminareviewer.com/guide/export-from-pubmed/`
- Scopus Export Guide: `https://luminareviewer.com/guide/export-from-scopus/`
- Web of Science Export Guide: `https://luminareviewer.com/guide/export-from-web-of-science/`
- Embase Export Guide: `https://luminareviewer.com/guide/export-from-embase/`
- CINAHL Export Guide: `https://luminareviewer.com/guide/export-from-cinahl/`
- Cochrane Library Export Guide: `https://luminareviewer.com/guide/export-from-cochrane-library/`
- Google Scholar Export Guide: `https://luminareviewer.com/guide/export-from-google-scholar/`
- PsycINFO Export Guide: `https://luminareviewer.com/guide/export-from-psycinfo/`

### Alternative & Competitor Comparisons
- Rayyan Alternative: `https://luminareviewer.com/compare/rayyan-alternative/`
- Covidence Alternative: `https://luminareviewer.com/compare/covidence-alternative/`
- ASReview Alternative: `https://luminareviewer.com/compare/asreview-alternative/`
- DistillerSR Alternative: `https://luminareviewer.com/compare/distillersr-alternative/`
- EndNote Alternative: `https://luminareviewer.com/compare/endnote-alternative/`
- Excel Alternative: `https://luminareviewer.com/compare/excel-alternative/`
- Elicit Alternative: `https://luminareviewer.com/compare/elicit-alternative/`

## Practical Citation Workflows & Reproducible Examples
- [PubMed NBIB to RIS: conversion and field checks](https://luminareviewer.com/guide/pubmed-nbib-to-ris/): Convert PubMed NBIB references to RIS, compare the fields, and download matched synthetic examples. Lumina also imports NBIB directly.
- [RIS import missing abstracts: diagnose and fix the export](https://luminareviewer.com/guide/ris-import-missing-abstracts/): Find out why abstracts are missing from RIS imports. Inspect AB tags, compare export settings, and test downloadable full and citation-only examples.
- [Import a Rayyan export into Lumina: citations and migration limits](https://luminareviewer.com/guide/import-rayyan-export/): Move Rayyan citation exports into Lumina using RIS. Check what transfers and preserve decisions, notes and the original audit log separately.
- [Deduplicate PubMed and Embase results: a worked example](https://luminareviewer.com/guide/deduplicate-pubmed-embase/): Follow a downloadable PubMed and Embase overlap example: four records, one duplicate, three unique citations, with PRISMA counting and matching limits.
- [Citation import benchmark: reproducible metadata checks](https://luminareviewer.com/guide/citation-import-benchmark/): Download Lumina’s synthetic NBIB and RIS fixtures and reproduce record counts, missing-abstract checks, migration limits and cross-database deduplication.
