Protocol
Extract with evidence and an unknown state
An empty field can be the correct answer. Define how to represent missing or uncertain information before asking AI to extract it. Then keep each important value tied to the source that actually contains it, rather than to what the model thinks probably belongs there.
Before you use it
- Use when
- You want structured records from material you are permitted to process with the chosen tool.
Keep this limit in mind
- Correct JSON is not correct data. Do not infer sensitive attributes or transfer restricted documents into an unapproved service.
An idea to explore.
Try this idea.
Inspect the example, try the idea in your situation, then check your next move.
Why it matters
An empty field can be the correct answer. Define how to represent missing or uncertain information before asking AI to extract it. Then keep each important value tied to the source that actually contains it, rather than to what the model thinks probably belongs there.
Steps
- Define fields, allowed values and an explicit null or unknown state.
- Require a source location or short supporting excerpt for important values.
- Validate the output shape and check a sample against the original.
- Review every high-impact or uncertain field and reject unsupported guesses.
An example
When no approval date appears in the source, return unknown. Do not turn the document's creation date into an approval date.
Check your result
The retained fields satisfy the schema and their important values are checked against the source.
Save and practise this idea
My learning
Read. Recall. Repeat.
Save this idea and return to it on any device.
Keep exploring
Connected ideas
You are readingExtract with evidence and an unknown state
Ideas that link here
These ideas link to this one. Check their conditions and limits before use.
Useful with this ideaFind the AI task boundaryUse before this idea
Find the AI task boundary
Evidence and sources
First-party source of this practical exercise.
Provenance for selected structured fields. The exercise is editorial practice; this link supplies no efficacy finding.
Extract a Table with AI, Then Sample-Check the Rows · data/life-os-content/ai-structured-extraction-check.json
All sources (1)
Edition and review
Version 0.1.0 · Published
Edition checked · 2026-11-12