> ## Documentation Index
> Fetch the complete documentation index at: https://www.doc-reviewer.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Load and evaluate your documentation web pages by URL

> Load one or more web pages by URL using a headless Chromium browser, then evaluate their instructions as a single document in Doc Reviewer.

Instead of uploading a file, you can point Doc Reviewer at a live URL. The app fetches the page using a headless Chromium browser, extracts instruction sections from the HTML, and treats the result as a regular document ready for evaluation. You can add multiple pages to the same document before running the evaluation.

<Warning>
  Web page evaluation requires Chromium, which is not included in the `.exe`. You must install it once on any machine where you want to use this feature — whether you are running the `.exe` or from source. Open a terminal and run:

  ```bash theme={null}
  py -3.11 -m playwright install chromium
  ```
</Warning>

## Load a web page

<Steps>
  <Step title="Open the Evaluation page">
    Click **Evaluation** in the sidebar, or navigate to any project and click **Evaluate by URL**.
  </Step>

  <Step title="Select the By URL tab">
    At the top of the Evaluation page, click the **By URL** tab to switch from file upload to URL mode.
  </Step>

  <Step title="Paste a URL and click Load">
    Paste the full URL of the page you want to evaluate (must start with `http://` or `https://`) and click **Load**. Doc Reviewer launches Chromium in headless mode and fetches the page.
  </Step>

  <Step title="Wait for the page to load">
    Chromium renders the page fully before extraction begins, so JavaScript-rendered content and single-page applications (SPAs) load correctly. Fetching typically takes a few seconds. A loading indicator is shown while Chromium is running.
  </Step>

  <Step title="Add more pages (optional)">
    After the first page loads successfully, a **+ Add page** button appears in the document header. Click it, paste another URL, and click **Load** to append that page's instructions to the same document. Repeat for each additional page you want to include.
  </Step>

  <Step title="Proceed to evaluation">
    Once all pages are loaded, the document tree on the left shows all extracted instruction sections. Review the sections, adjust classifications if needed, and click **Evaluate** to start the LLM evaluation.
  </Step>
</Steps>

<Note>
  All pages you add become part of a single document. Each instruction block extracted from each page appears as a separate section in the document tree and is evaluated individually.
</Note>

## How web parsing works

Doc Reviewer uses two parsing strategies depending on the source site:

<Tabs>
  <Tab title="Positive Technologies web help">
    When the page contains custom `<instruction>`, `<action>`, or `<task>` tags — the structure used in Positive Technologies web help — Doc Reviewer uses a specialized parser that:

    * Treats each `<instruction>` block as a separate document section
    * Correctly numbers steps inside `<action>` tags
    * Strips layout artifacts such as soft-hyphen characters (`&shy;`)

    This parser produces the highest-quality extraction and is the primary use case for web evaluation.
  </Tab>

  <Tab title="Other websites">
    For any site that does not use the Positive Technologies web help structure, Doc Reviewer falls back to a generic HTML-to-Markdown converter. The entire page content is converted to Markdown and then split into sections.

    Extraction quality varies by site. Heavily styled pages, tab-based navigation, and content hidden behind JavaScript interactions may produce incomplete or noisy sections.
  </Tab>
</Tabs>

<Note>
  Because Chromium renders JavaScript before extraction, the app supports JS-rendered pages and SPAs. Static HTML pages load faster, but both work correctly.
</Note>