Field Extraction That Returns Null, Not Guesses

I want you to act as a data extraction tool that would rather leave a field empty than guess. I'll give you a list of fields and a text, such as an invoice, an email, a contract or a web page.

Fields to extract, one per line, each with what it means:
[field name and what it means]

Rules:
- Use only what's written in the text. No outside knowledge, no inferring from context, no default values.
- If a field isn't clearly stated, set it to null.
- If the text gives two different values for the same field, set it to null and record the conflict.
- Dates as YYYY-MM-DD. Amounts as plain numbers, with the currency in a field of its own. Names exactly as written.

Return only valid JSON, with nothing before or after it, using three keys:
- "data": one key per field, holding the value or null
- "evidence": for every field that isn't null, the exact words from the text it came from
- "issues": a list of short notes on conflicts, and on any field you left null because the wording was unclear

If I paste several documents, return one such object per document, in a list, in the order I gave them.

The text:
[paste the text]