You have opened a webpage using the browser tool and taken a snapshot. Your task is to extract all text content from this page that is relevant to the research query below.

Research Query: {{query}}
Max output length: {{max_chars_per_source}} characters

---

EXTRACTION RULES:

1. **Include**: Main article body, key facts, data points, quotes, dates, names, and any content that directly or indirectly addresses the query.

2. **Exclude**: Navigation menus, cookie banners, subscription prompts, footer content, social sharing buttons, ad copy, and any boilerplate UI text.

3. **Format**: Return plain text only — no markdown formatting, no headers, no bullet points. Just the extracted content as flowing prose or structured data as it appears on the page.

4. **Relevance filter**: If a paragraph has zero relation to the query, skip it. You are curating, not transcribing.

5. **Truncation**: If the relevant content exceeds {{max_chars_per_source}} characters, keep the highest-signal portions (typically: opening paragraphs, key data tables, conclusions). Note at the end: "[Truncated — original page has more content]"

6. **If the page is empty or unreadable** (CAPTCHA, login wall, error page): Return exactly this string: "FETCH_FAILED: Page could not be read — [brief reason]"

---

Return only the extracted text. No preamble, no explanation, no metadata. Start directly with the content.
