AI Skill Report Card
Cataloguing Archival Botanical Plates
Quick Start13 / 15
Given raw scraped text from a digital collections page, extract and normalize into a standard archival record block:
Title: [as given, preserve original spelling/accents]
Author/Creator: [name]
Role: [Artist/Engraver/Author/Photographer/etc.]
Record Type: [Digitised Book Plate / Manuscript / Photograph / etc.]
Collection: [collection name]
Reference: [call number/identifier]
Date: [date or date range]
Scope & Content: [descriptive paragraph]
Extent: [physical extent, e.g. "1 item"]
Is Part Of: [parent publication/series]
Subjects: [subject terms]
Repository: [holding institution]
Copyright: [rights statement]
Credit Line: [citation format]
Usage Terms: [license/terms]
Strip all UI chrome (search filters, nav buttons, field-picker lists) — keep only the actual record data.
Recommendation▾
Add a second example showing a messier or edge-case input (e.g. missing fields, OCR garbage, non-botanical item) to demonstrate robustness beyond the single happy-path example
Workflow13 / 15
Progress:
- Step 1: Identify and discard UI/navigation noise (search forms, field dropdowns, "Next/Previous" buttons, filter labels)
- Step 2: Locate the "Information" block — this contains the actual record
- Step 3: Map each label-value pair to standard schema fields
- Step 4: Normalize dates into ISO-friendly ranges where possible (e.g. "1807-1816")
- Step 5: Preserve original-language titles verbatim, but note any obvious OCR/encoding errors (e.g. "Elancée" rendered as "Elancee")
- Step 6: Flag missing standard fields as
Not statedrather than omitting them - Step 7: Output clean structured record
Recommendation▾
Include an example of 'bad' output (what NOT to produce) to contrast with the good normalized record, since the grading rubric rewards showing both good and bad outcomes
Examples14 / 20
Example 1:
Input:
Raw scraped page with search UI dropdown (hundreds of field names) followed by:
191. Phalangium Elatum. Phalangere Elancée Information Title 191. Phalangium
Elatum. Phalangere Elancee Author or creator Langlois Role Engraver Record
type Digitised Book Plate Collection Original Reference BP/49371-4001/11
Date 1807-1816 Scope & content Colour-printed engraved plate depicting
Phalangium elatum...
Output:
Title: 191. Phalangium Elatum. Phalangère Élancée
Author/Creator: Langlois
Role: Engraver
Record Type: Digitised Book Plate
Collection: Not stated
Reference: BP/49371-4001/11
Date: 1807–1816
Scope & Content: Colour-printed engraved plate depicting Phalangium elatum.
The book plate bears the imprints 'P.J. Redouté pinx' and 'Langlois sculp'.
From the book 'Les Liliacées, Vol 4' by Pierre Joseph Redouté.
Extent: 1 item
Is Part Of: Les Liliacées, Vol 4
Subjects: Portrait format; French
Repository: Royal Horticultural Society Lindley Library
Copyright: No known copyright restrictions
Credit Line: RHS Lindley Collections
Usage Terms: Public Domain
Recommendation▾
Expand completeness by addressing edge cases like multiple creators/roles, ambiguous date ranges, or records with no Scope & Content at all
Best Practices
- Treat "Original Reference" and "Reference" as the shelf mark/identifier field — map to
Reference, notCollection, unless a distinct collection name is also given. - Keep Scope & Content as a single coherent descriptive paragraph; don't fragment it.
- Reconstruct accented characters in titles when the artwork's subject/language makes the correct form obvious (e.g. French botanical plate titles).
- When multiple near-duplicate field labels appear (common in exported schemas, e.g. "Description" listed three times), only use the one with an actual populated value.
- Always surface rights/usage information clearly — this is often the most reused part of the record.
Common Pitfalls
- Don't include the giant field-name/search-filter list in output — it's UI scaffolding, not data.
- Don't confuse "Author Institution" or "Firm name" boilerplate labels with actual populated values if absent from the record body.
- Don't drop diacritics or silently "correct" the title without noting the original scraped form if it's ambiguous.
- Don't invent a Collection name if only an Original Reference/shelf mark is present.