Evidence architecture and provenance · Lesson 4

Event-centred relationship modelling

5 min read

Lesson objectives

  • Model people, events and objects as linked records; produce an event-centred data map.
Reading tools

A conventional catalogue may have one field reading “the vessel came from workshop S, then belonged to family P, then entered the museum.” The sentence does not reveal whether these parties and dates rest on independent records or on a repeated summary. Event-centred modelling breaks it into an entity, events and relationships, each with a source. The entity is the vessel identified by its mark and description. Events may include manufacture, transfer, photography or entry. Relationships connect those events to participants, times and places. Some values can remain unknown. The place of discovery must not silently become the place of manufacture. The aim is to make the reasoning trail open to challenge when a new source appears.

CIDOC CRM provides a conceptual model for events and relationships in cultural heritage information. This lesson does not require memorising all its classes or claiming that every institution must adopt it. Its practical value is distinguishing “the vessel was made at workshop S” from “the donor said that the vessel was made at workshop S.” The former asserts a historical production event; the latter records a documentation event and its speaker. In the absence of further evidence, record the donor's statement and link it to the manufacture claim as testimony. Do not populate a confirmed-origin field with the workshop. A simple table can preserve this difference without a graph database.

Consider a fictional dossier for copper bowl N-21. A 2010 intake form says a donor delivered it to the institution. Her 2009 letter says her grandmother bought it from a market seller around 1950. A clear “H” mark appears on its surface, but no record identifies the mark's maker. At minimum, we need separate entities for the bowl, intake form and letter, as well as the donor and earlier holder as people named in records rather than makers. Record the 2010 entry event and connect it to the donor and intake form. Record an approximate 1950 purchase claim sourced to the letter, phrased “according to the donor's account.” Leave the maker and manufacture date unknown. The mark is an observable feature, not enough by itself to decode the maker's name.

Start the graph from a user's question rather than every field in a standard. If a researcher asks “which materials support the story of this instrument?”, the necessary relationships are instrument identifier, associated images and recordings, the event or statement connecting them, and the source for the link. If a collections manager needs to inspect movement, add transfer events, custodians and their dates. Fewer unsupported relationships make a better model than many arrows. With conflicting sources, represent two claims and their conflict status; one arrow should not conceal an editorial preference. Explicit unknowns are useful: “performance date unknown,” together with the last search made for it.

Before approval, ask of every edge whether a source states it, whether it is an inference from specified clues, or whether it is only a hypothesis to test. Label the relationship accordingly. If an identifier link changes after a transcription error is found, keep a change log: old and new identifiers, reason for correction, editor, and any public text that depends on the link. This makes the graph a stewardship tool rather than an attractive picture. Technical equality of identifiers cannot by itself establish the truth of a narrative or the right to publish the material. Access is a separate decision developed in the next module.

Test a miniature graph for bowl N-21. The edge “donor — delivered — bowl” is supported by the 2010 form and describes a limited entry event. “Grandmother — bought — bowl” is supported by the 2009 letter, but remains a reported, approximate purchase. “Mark H — identifies — maker” has no support and should be a research question, not a confirmed edge: is there a documented register of makers' marks to compare? These three relationships share a bowl but differ in status and event type. If the database cannot express “reported” or “under investigation,” keep a separate claim record rather than forcing assertions into confirmed fields. A public label should draw only on the status each edge permits.

Now change one assumption: the 2009 letter turns out to concern a different family bowl. Only the purchase edge and any description depending on it should change. The 2010 intake and observed mark remain. If a model cannot support that limited correction, it has merged independent claims too tightly. Its value is measured by the team's ability to answer a professional question and repair an error, not by the number of nodes it contains.

Lesson summary

A conventional catalogue may have one field reading “the vessel came from workshop S, then belonged to family P, then entered the museum.” The sentence does not reveal whether these parties and dates rest on independent records or on a repeated summary