README.md

metadata

title: EvidenceGraph Research Intelligence
emoji: 🕸️
colorFrom: green
colorTo: blue
sdk: static
app_file: index.html
short_description: Trace publications, policy claims and evidence networks.
license: mit
pinned: false

EvidenceGraph Research Intelligence

An open-source, browser-local evidence-network studio for academic and policy research.

Author / portfolio: Rodrigo Pantoja · Hugging Face @Hugdrigo
License: MIT · App architecture: Static HTML/CSS/ES modules · Hosting: Free Static Space

What it does

• Build a provenance-aware network of four node types: publications, claims, quantitative indicators, and policy hypotheses.
• Add supports, challenges, or context links from researcher-entered excerpts with explicit source locators.
• Inspect and trace graph connections, including shortest paths and relationships.
• Retrieve publication metadata from the public Crossref API, review the result, and add it to your research library.
• Retrieve the latest observed value and reporting year for selected World Bank indicators.
• Compare competing hypotheses without automatic truth claims.
• Audit documentation completeness, conflicting link labels, and unconnected records.
• Export a complete, restorable JSON session, node/edge CSV files, GraphML, and a Markdown research report.

Important: There is no generative AI, automatic scientific truth checker, or independent validation of user-entered excerpts. Bibliographic metadata does not establish a study's results. Researcher-labeled relations express hypotheses or judgments, not causal proof. Graph path proximity does not establish entailment. World Bank data provides numeric context, not a causal estimate.

Quick start

1. Open the public App page.
2. The initial demonstration is 100% synthetic, including fictional publications, passages, and indicator values.
3. Select 01 Evidence network to inspect nodes and link colors.
4. Open 02 Evidence desk to add a research claim and a located excerpt. Add real sources under 03 Literature hub or 04 Indicator hub first.
5. Use 06 Quality audit to find weak documentation.
6. Go to 07 Export & methods and download Complete session (.json) to save your work.
7. To start genuine research, click New empty session in the Evidence network first. Otherwise the demo's fictional nodes remain mixed with your own records and the DEMO/MIXED badge stays visible.

External data, and what it is not

Crossref (https://api.crossref.org/works): fetches titles, authors, DOI, year and venue. Results may contain mistakes or omissions. Full-text articles are not retrieved. Searches send a query to Crossref from the browser.

World Bank Indicators API v2 (https://api.worldbank.org/v2): retrieves the latest available non-null annual observation from a selected country and indicator within the app's request date range. The actual observation year is always stored. Some indicators lack recent data. Searches send country and indicator codes to World Bank.

Both integrations depend on client internet connectivity and the provider's availability/CORS restrictions. An offline fictional example works without any API. Browser API behavior must be verified after live deployment.

Reproducibility and source files

File	Purpose
index.html	Accessible application entry point
styles.css	Responsive interface and network visualization styles
app.js	Navigation, interaction, metadata retrieval and exports
engine.js	Deterministic graph, audit, parsing and export functions
demo.js	Fictional example with explicit synthetic provenance
TEST_CORE.mjs	node --test TEST_CORE.mjs deterministic checks
TEST_BROWSER.py	Optional local Playwright checks for the rendered interface
DEMO_SESSION.json	Portable copy of the fictional demonstration dataset
Documentation .md files	Methodology, limitations, evidence protocol and deployment

Local tests: run node --test TEST_CORE.mjs. Browser tests require Python Playwright and Chromium; see QA_REPORT.md. The application itself requires no installation and no dependencies.

Security and privacy

No database, login, analytics pixel, tracker, or server-side storage is included. Sessions live in browser memory and disappear upon refresh unless exported. Untrusted titles and metadata are escaped before rendering; external links are checked for HTTP(S). JSON import has size and graph-structure limits. Do not include sensitive or unpublished research in a public Git repository. Crossref / World Bank searches necessarily transmit request parameters to those providers.

Limitations

Audit scores measure whether evidence documentation is present, not source validity, independence, journal quality, impact, fairness, or factual correctness. Publication duplication is DOI-based, not semantic. The network visualization is optimized for small projects. This prototype is suitable for a public methods demonstration and exploratory research organization, not automated policy decisions or systematic review certification.

See METHODOLOGY.md, EVIDENCE_PROTOCOL.md, LIMITATIONS.md, and USER_GUIDE.md.