Provenance Data Model and Query Patterns

Reference material for search_provenance. Consult this page when you need to interpret raw provenance text, understand the data model, or construct specialised query patterns.

Provenance Text Format (AAM Standard)

The Rijksmuseum's provenance text follows the AAM (American Alliance of Museums) standard for provenance notation:

• Semicolon-delimited events: each ownership change is separated by ; — Owner A, City, date; Owner B, City, date
• Chronological order: earliest known owner first, current holder last
• Bare names = ownership: a name without a transfer keyword (e.g. Pieter Six (1618–1707), Amsterdam) indicates an ownership period — classified as collection
• Ellipsis = gap: … or ... marks a gap in the documented chain
• Curly braces = citations: {Bredius 1935, p. 47} are bibliographic references, not provenance events
• Question mark = uncertainty: ? sale, Amsterdam, 1804 — the transfer or attribution is uncertain
• Cross-references: (see also SK-A-3137) links to pendant or companion works that share provenance
• Life dates: Name (1650–1720) — dates in parentheses after a name are birth–death years, not event dates

The Rijksmuseum extends the AAM standard with Dutch-language keywords (schenking = gift, bruikleen = loan, verwerving = acquisition, aangekocht = purchased) and institutional conventions (Inv. = inventory reference, L. = Lugt collector mark number).

Transfer Types (CMOA/PLOD-aligned Vocabulary)

Transfer types follow the CMOA Art Tracks thesaurus (Carnegie Museum of Art), aligned with the PLOD framework (Art Institute of Chicago).

Extensions beyond CMOA: recuperation (physical recovery vs legal return), collection (AAM bare-name convention), inventory (attestation events).

Event Flags

Inheritance granularity: use by_descent for works inherited by named relatives, widowhood for widow/widower inheritance specifically, or inheritance for the generic case. To catch all inheritance-related transfers: transferType: ["by_descent", "widowhood", "inheritance"].

Party Roles and Positions

Each party in a provenance event has a role (what they did) and a position (their side of the transfer):

Positions (sender/receiver/agent) are derived from roles via deterministic mapping, with LLM enrichment for ambiguous cases. The positionMethod field tracks how each party's position was determined: role_mapping (deterministic), llm_enrichment (LLM-classified), or llm_disambiguation (LLM-resolved merged party text).

Coverage: ~86K parties extracted across ~101K events. Not all events have named parties — bare-name collection events and cross-references often lack structured party data.

Date Representation

Dates use qualified single years with temporal bounds, similar to the EDTF (Extended Date/Time Format) approach:

Historical Currencies

Prices are stored in their original historical currency — no inflation adjustment or cross-currency conversion is performed. Supported currency values:

guilders, pounds, francs, livres, napoléons, guineas, belgian_francs, deutschmarks, reichsmarks, swiss_francs, euros, dollars, yen, marks, louis_d_or.

Pre-decimal notations (£.s.d, fl. X:Y:-) are converted to decimal equivalents of the base currency unit.

Note on batch prices: en bloc prices (a batch sold together, e.g. the entire Mannheimer collection for fl. 6,350,000) are attributed to every individual item in the batch. These events are flagged with batchPrice: true — always filter them out when ranking or comparing prices.

Enrichment Provenance

Every record carries provenance-of-provenance metadata tracking how it was determined:

• parseMethod: how the event was parsed (peg, regex_fallback, cross_ref, credit_line)
• categoryMethod: how the transfer type/category was determined (type_mapping = parser, llm_enrichment = LLM, rule:transfer_is_ownership = validated rule)
• positionMethod (on parties): how the party position was determined (role_mapping = parser, type_mapping = from transfer type, llm_enrichment = LLM, llm_disambiguation = LLM-decomposed merged text)
• enrichmentReasoning: the LLM's reasoning for any non-deterministic decision

When results contain LLM-enriched records, search_provenance provides a URL to an enrichment review page showing the full methodology and reasoning for each decision. Always show this URL to the user.

Querying by enrichment method

# Artworks where transfer type was classified by LLM (170 events)
search_provenance(categoryMethod="llm_enrichment", maxResults=10)

# Artworks where party position was assigned by LLM
search_provenance(positionMethod="llm_enrichment", maxResults=10)

# Artworks where merged party text was decomposed by LLM (204 splits)
search_provenance(positionMethod="llm_disambiguation", maxResults=10)

For collection-wide distribution of methods:

collection_stats(dimension="categoryMethod")
collection_stats(dimension="positionMethod")
collection_stats(dimension="parseMethod")

Parse Method Values

The parseMethod field records how each provenance record was processed:

credit_line events are particularly useful: they recover acquisition context (donor name, purchase fund) that the provenance text omits.

Provenance Facets

search_provenance supports facets: true, returning 5 facet dimensions alongside chain results:

• transferType: distribution of transfer types in matching events
• decade: temporal distribution of matching events
• location: geographic distribution (city-level, top 20)
• transferCategory: ownership vs custody vs ambiguous
• partyPosition: sender vs receiver vs agent

All entries include count and percentage.

# Faceted overview of wartime confiscations
search_provenance(transferType="confiscation", dateFrom=1933, dateTo=1945, facets=true)

# Faceted overview of a dealer's activity
search_provenance(party="Goudstikker", facets=true)

# Faceted audit of LLM-classified events
search_provenance(categoryMethod="llm_enrichment", facets=true)

Tested Query Patterns

Collector profiling

# All works associated with a collector across all event types
search_provenance(party="Mannheimer", maxResults=50)

# Collector as seller specifically
search_provenance(party="Goupil", maxResults=20)
# Then filter results: role="seller" + position="sender" to map direction of trade

# Ownership durations for a family name
search_provenance(layer="periods", ownerName="Six",
                  sortBy="duration", sortOrder="desc", maxResults=20)

Acquisition channel analysis

# creditLine covers ~358K artworks — far more than parsed provenance (~48K)
search_artwork(creditLine="Drucker-Fraser", compact=true)
search_artwork(creditLine="Vereniging Rembrandt", type="painting", compact=true)

Wartime provenance

# Anti-join: confiscated but never restituted
search_provenance(transferType="confiscation",
                  excludeTransferType="restitution", maxResults=20)

# Works with documented gaps + events in the wartime period
search_provenance(hasGap=true, creator="Rembrandt",
                  dateFrom=1933, dateTo=1945, maxResults=20)

# Recuperation events (Allied recovery, distinct from legal restitution)
search_provenance(transferType="recuperation", maxResults=20)

Price and market history

# Most expensive recorded transactions in guilders
# Check batchPrice on results — true means en bloc total, not individual price
search_provenance(hasPrice=true, currency="guilders",
                  sortBy="price", sortOrder="desc", maxResults=20)

# Price history for a specific work
search_provenance(objectNumber="SK-A-2344", layer="events")
# Check unsold flag: unsold lots have prices but no sale occurred

Multi-generation family collections

search_provenance(transferType=["by_descent", "widowhood"],
                  layer="periods", minDuration=50,
                  sortBy="duration", sortOrder="desc", maxResults=20)

Chronological exploration

# Earliest documented provenance events in the collection
search_provenance(dateFrom=1400, dateTo=1500, sortBy="dateYear",
                  sortOrder="asc", maxResults=20)

Decade-level time series

Use collection_stats for single-call time series — no manual pagination loop needed:

# Sale events per decade 1600–1900
collection_stats(dimension="provenanceDecade", transferType="sale",
                 dateFrom=1600, dateTo=1900)

# Half-century bins
collection_stats(dimension="provenanceDecade", transferType="sale",
                 dateFrom=1600, dateTo=1900, binWidth=50)

# Confiscation events by decade (wartime distribution)
collection_stats(dimension="provenanceDecade", transferType="confiscation")
# Reveals 1790s + 1940s bimodal pattern (French Revolution + WWII)