§6.6 · BUILDING · 4 MIN READ · EDITION 1.0
Handling scanned and unstructured evidence
Scanned drawings, handwritten inspection sheets and unindexed PDF archives are common in wind quality work; treat OCR and layout-extraction quality as its own evaluated step, not an invisible preprocessing detail.
A capability's downstream performance is bounded by the quality of text and structure extracted upstream; a perfect model reading garbled OCR output will still produce a wrong answer, and the error will look like a model failure.
Where scanned evidence is common, budget for a dedicated readability project — rescanning, indexing, structuring — before expecting an AI-supported step to perform reliably on it.
PUT THIS INTO PRACTICE
- 01Evaluate OCR and extraction quality separately from the end task.
- 02Treat unreadable evidence as a data-readiness gap, not a model limitation.
- 03Prioritise readability improvement where scanned evidence is common.
RELATED INSTRUMENTS
CITE THIS SECTION
§6.6 · Edition 1.0 · Handling scanned and unstructured evidence