Cookbook

This is a photo of a project-discussion whiteboard. Please:
1. Extract the main topics and conclusions on the whiteboard.
2. Organize them into a project plan with goals, key milestones, risks, and items needing confirmation.
3. Generate a task list with fields: task, owner (null if not determinable), priority, dependencies, suggested due date.
4. Separately list anything that is unclear or needs human confirmation.

## Project plan
## Task list
## Risks and dependencies
## Items to confirm

Extract data from this chart and return it as JSON:
{
  "chart_type": "",
  "title": "",
  "x_axis": "",
  "y_axis": "",
  "series": [
    {
      "name": "",
      "points": [
        {"label": "", "value": null, "confidence": "high|medium|low"}
      ]
    }
  ],
  "insights": [],
  "uncertain_fields": []
}

Requirements:
- If a value can only be estimated, mark confidence as low or medium.
- Do not fabricate information that isn't in the chart.
- If axes, units, or legend are unclear, put them in uncertain_fields.

Extract structured information from this receipt and return it as JSON:
{
  "merchant": "",
  "date": "",
  "currency": "",
  "total_amount": null,
  "tax_amount": null,
  "items": [
    {
      "name": "",
      "quantity": null,
      "unit_price": null,
      "amount": null
    }
  ],
  "payment_method": "",
  "uncertain_fields": []
}

Requirements:
- Amounts must come from the receipt itself; do not infer.
- Fields that are unclear should be null and written into uncertain_fields.
- Preserve the original currency and date format.

Output the line items in CSV with columns:
merchant,date,item_name,quantity,unit_price,amount,currency,confidence

This is a screenshot of a web page. Use React + Tailwind CSS to recreate it.

Requirements:
1. First describe the page structure, layout, and main visual elements.
2. Then produce runnable React component code.
3. Use semantic naming; don't depend on real-world brand assets from the screenshot.
4. For images and icons you can't determine, use placeholder elements.
5. Keep the layout sensible on both mobile and desktop.

This is a screen recording of someone using a piece of software. Analyze:
1. What actions did the user take, in order?
2. At which step did the anomaly begin?
3. What does the anomaly look like?
4. What are the likely causes, ordered by probability?
5. Suggested investigation steps and likely fixes.

Output in Markdown, and put anything unclear into a separate "Info needed" section.

## Action timeline
## Anomaly point
## Likely causes
## Investigation steps
## Info needed

Compare these images and output:
1. Similarities.
2. Differences, grouped by visual layout, text content, data values, and state changes.
3. Possible impact.
4. Differences requiring human review.

If a difference can't be confirmed, say so explicitly — don't guess.

Return valid JSON only — no Markdown.
If a field can't be confirmed from the image, set it to null.
For fields with uncertain recognition, add the field name and reason to uncertain_fields.

{
  "field": "total_amount",
  "value": 128.5,
  "confidence": "medium",
  "evidence": "The 'Total' line at the bottom of the receipt"
}