LEVEL 02 · LESSON 15
Metadata
Metadata tells a system which information applies, where it came from, and who may use it.
What it is
Metadata is descriptive information attached to content: its source, owner, date, region, document type, section, or access classification.
Why it matters
Similarity alone cannot tell you whether a passage is current, authorized, or applicable to the customer in front of you.
A business example
A support assistant finds refund policies for several countries. Filtering by country, product, and effective date prevents an otherwise relevant policy from being applied to the wrong order.
How it actually works
Capture metadata during ingestion and carry it through retrieval into citations and logs. Use stable identifiers to connect chunks back to documents and sections. Some fields narrow a search before results are selected; others explain the result afterward. Access-control fields must come from trusted systems and be enforced by application logic. Metadata needs an owner and update process, because stale labels can silently produce wrong answers.
A common misconception
Putting ‘confidential’ in a text field does not enforce confidentiality. The retrieval system must actually check the user’s permissions.
ONE SENTENCE TO REMEMBER
Metadata tells a system which information applies, where it came from, and who may use it.
CHECK YOUR UNDERSTANDING
Chunks from HR documents are labeled “confidential” in a metadata field. What still has to be true for confidential content to stay protected?
Show the answer
Answer: B.
A label does not enforce anything by itself. Access-control fields must come from trusted systems and be enforced by application logic during retrieval, so unauthorized content never reaches the model’s context.
Take it into your next meeting
Which three labels would prevent the most retrieval mistakes in your organization?
One more concept, understood.
Mark this lesson complete to keep your place.