# Beginner user guide

## What is an evidence graph?

Imagine a wall with cards connected by string. One card is a paper, another is a policy idea, and the string says *why* they may be related. Here the strings can say **supports**, **challenges**, or **context**. Every string should contain the real words or table location you checked.

## First test (fictional data)

1. Open the App tab. You should see **EvidenceGraph Research Intelligence** and the banner **SYNTHETIC DEMO / MIXED SESSION**.
2. Open **01 Evidence network** and click a blue publication node.
3. Scroll to see its details and connected excerpts. The title explicitly says fictional.
4. Try **Trace a research connection** by selecting two cards and pressing the button.
5. Open **05 Hypotheses** and observe supporting, challenging and context link counts.
6. Open **06 Quality audit** to see records that need human attention.
7. Open **07 Export & methods** and download a JSON session, CSV and Markdown report.

## Start a real research project

1. On **01 Evidence network**, click **New empty session**. This removes the fictitious demo *from the current browser tab only*. It never deletes your Hugging Face repository.
2. Use **03 Literature hub** to search Crossref and import bibliography details, or enter a publication manually.
3. Use **04 Indicator hub** to request an official numeric indicator and preserve the observed year.
4. Use **02 Evidence desk** to write an original claim/hypothesis and enter a source excerpt plus a page or table locator.
5. Compare hypotheses and check audit warnings. **Do not use unverified references to claim scientific validity.**
6. Export the complete session as JSON before refreshing or closing the tab.

## Restore a session

Go to **07 Export & methods → Restore a saved session**, select the JSON exported earlier, and confirm replacement. If the imported file is malformed, the existing session remains intact.

## No-cost hosting

Static Hosting does not execute Python or deploy an LLM. The analysis here is a browser-only combination of graph algorithms, metadata fetching, user-authored links, and deterministic documentation diagnostics.