## 1. Check identity, dates and material type

Read source-data.json only. Check 8 unique ids, dates, empty text and URLs; preserve wording. Synthetic URLs must be empty. Save intake.md; do not read X, delegate or build a page yet.

Expected: An eight-item intake report.
Check: No fabricated X link or real-person attribution for synthetic text.

## 2. Connect themes to evidence

Create 2–4 themes with id, title, observation and evidence array. Cite existing demo ids only. Infer no occupation, income or health and give no percentage scores. Deliver claims.json.

Expected: A supported theme table.
Check: Every evidence id exists; no claim is stronger than its sources.

## 3. Agree content order and mobile layout

Write site-brief.md from reviewed claims: introduction, theme order, evidence ids, source area and a single-column 375px layout. Label synthetic data; add no name, contact details or achievements. Deliver design notes only.

Expected: An implementable content/layout brief.
Check: Each card links to evidence; the synthetic label is visible near the top.

## 4. Generate a standalone static page

Read input, claims and brief; generate standalone index.html with no external scripts, fonts or images and no dependencies. Use real same-page ids and retain all source text. One Bot works sequentially; do not delegate, deploy or send messages.

Expected: A standalone HTML file and its actual path.
Check: Themes match sources and no evidence target is invented.

## 5. Check actual desktop, mobile and evidence links

Open the delivered page in an available browser. Check desktop and 375px: all evidence links, unchanged text, no invented facts or whole-page horizontal overflow. Return actual screenshots and checks. If unable to open, mark browser checks unverified; do not simulate screenshots.

Expected: Actual screenshots and a checklist.
Check: Evidence comes from the real page and states untested areas.

## 6. Revise evidence before updating the page

For <theme id/revision reason>, update claims, then brief and HTML. Preserve the prior version and change notes; recheck evidence and mobile layout. Confirm scope separately for real inputs. Do not inherit sample conclusions, deploy or delegate.

Expected: A consistent revision and recheck.
Check: All deliverables agree; no invented posts or hidden historical limits.