Knowledge quality

A knowledge assistant cannot repair an unmanaged source collection by itself

Search and generation quality depend on readable files, current versions, meaningful metadata and people who own the content.

Reviewed

Quality dimensions

Prepare the source before evaluating the answer

Extraction
Scans, tables, images and unusual layouts may require OCR or document-specific processing.
Version
Current and superseded files need explicit status and replacement relationships.
Metadata
Owner, date, type, audience and sensitivity improve selection and lifecycle control.
Authority
The system needs rules for which source wins when documents disagree.

Lifecycle

Add, test, replace and remove

  1. 1

    Add

    Preserve the original, extract text and assign metadata.

  2. 2

    Test

    Use representative questions to verify document and passage retrieval.

  3. 3

    Maintain

    Replace outdated sources and verify deletion from retrieval.