Appearance
Modelling a heritage item in Wikidata means describing one real-world entity as an item with a clear P31 (instance of) class, then attaching sourced statements for date, place, creator, material and collection. Start from the class downward: decide what the thing is, give it identity statements, then add the descriptive and relational claims. Good modelling is reuse — copy the statement pattern of well-formed items of the same type rather than inventing your own.
How do I decide what counts as one item?
The unit is the entity, not the catalogue row. A bound volume of forty letters can be one item (the volume) with forty P361-linked letter items if each letter is individually significant. A coin hoard is one item if it is treated as a single find, or many if catalogued individually.
Ask: would a researcher ever want to query or cite this on its own? If yes, it deserves an item. If it only exists as a part of something else, model it as a qualifier or a sub-item.
What is the core statement skeleton for an object?
Every well-formed heritage item shares a backbone. For a medieval seal matrix:
text
P31 Q1424694 # instance of: seal matrix
P186 Q39782 # made from material: copper alloy
P571 +1300-00-00T00:00:00Z/8 # inception, decade precision
P170 Q... # creator (if known)
P195 Q... # collection
P217 "SM-0421" # inventory number (qualifier on P195)
P1071 Q... # location of creationLead with identity (P31, material, date), then provenance (creator, place), then institutional context (collection, inventory). Add a reference block to every line that is not self-evident.
How do I model works versus physical carriers?
This trips up newcomers constantly. A text and the manuscript that carries it are different entities:
| Concept | Item type | Key property |
|---|---|---|
| The intellectual work | P31 = literary work | links to author via P50 |
| The specific manuscript | P31 = manuscript | P629 edition/translation of -> work |
| The digital surrogate | usually not a separate item | use P4765 image URL on the object |
Keep the surrogate's metadata (resolution, IIIF manifest) attached to the physical item rather than spawning a third item, unless the digital object is itself notable.
How do I express uncertainty honestly?
Heritage data is full of "circa", "attributed to" and "possibly". Wikidata handles this with qualifiers rather than vague labels:
- Date ranges:
P1319(earliest date) andP1326(latest date). - Disputed attribution:
P1480(sourcing circumstances) = "attributed to". - Deprecated rank for claims now believed wrong, kept for transparency.
sparql
# Find items dated to a range rather than a fixed year
SELECT ?item ?earliest ?latest WHERE {
?item wdt:P195 wd:Q... ;
p:P571 [ pq:P1319 ?earliest ; pq:P1326 ?latest ] .
}Why model relationships, not just attributes?
The value of Wikidata over a local spreadsheet is the graph. Linking a manuscript to its scribe, the scribe to a scriptorium, and the scriptorium to a place lets a single query surface "all manuscripts produced within 50km of York before 1200". Flat catalogues cannot do that. Invest in P170, P1071, P361 and P195 links — they are what make the data analytically powerful.
How do I keep a collection internally consistent?
Define a small modelling profile before you start: which P31 classes you allow, required properties, and your date conventions. Document it on a project page and validate periodically with a SPARQL query that flags items missing required statements. Consistency across 500 items beats perfection on 5.
Key Takeaways
- The unit of modelling is the entity; if it is queryable or citable alone, give it an item.
- Lead with identity (
P31, material, date), then provenance, then collection context. - Separate intellectual works from their physical carriers using
P629. - Encode uncertainty with qualifiers and ranks, never with vague labels.
- Model relationships, not just attributes — the graph is the payoff.
- Write a short modelling profile and validate consistency with SPARQL.
Frequently Asked Questions
What property says what kind of thing a heritage item is?
Use P31 (instance of) for the class, such as Q87167 manuscript or Q860861 sculpture. Use P279 (subclass of) only when you are defining a class, not an individual object.
How do I link an item to its parent collection?
Use P195 (collection) to point to the holding institution's collection item, and P217 (inventory number) for the local accession ID. P361 (part of) handles intellectual sub-parts.
Should I model the physical object or the abstract work?
For unique heritage objects, model the object itself. For works that exist in many copies, separate the work item from each manuscript or edition item and link them with P629 (edition or translation of).
How precise should dates be?
Set the precision to match your evidence: year, decade or century. Use qualifiers P1319 (earliest date) and P1326 (latest date) when you only have a range.
What do I do when no exact property exists?
Search the property catalogue and existing similar items first. If nothing fits, raise a property proposal rather than forcing data into an unrelated property.
How do I record uncertainty about a creator?
Add P170 (creator) with the qualifier P1480 (sourcing circumstances) set to Q18122778 (attributed to) and attach a reference explaining the attribution.