TfP Docs

Museum Provenance

A provenance Campaign coordinates research into the ownership and movement of artworks or cultural objects.

This stresses a different part of TfP: large quantities of documentary evidence, conflicting hypotheses, expert review, and investigations that recursively expose new questions.

flowchart TD
    C[Catalogs / archives / auction records] --> D[Discover provenance gaps]
    D --> P[Standing provenance Problem]
    P --> A[Research Attempts]
    A --> R[Structured provenance Artifact]
    R --> V[Curator / agent review]
    V --> T[Trusted provenance result]
    T --> N[New investigative Problems]

1. The community defines a provenance standard

A museum consortium or research organization publishes a Skill:

@museum/provenance-research

It may contain:

SKILL.md
schemas/provenance-record.schema.json
examples/
source-quality guidance/
name-normalization tooling/
date/place normalization/

The structured record could include:

object identity

known custody events

date ranges

owners / holders

locations

source citations

certainty

conflicting evidence

unresolved gaps

candidate related archives

hypotheses

TfP does not need a ProvenanceRecord core entity.

It is an Artifact standard defined by the community.

2. Find gaps worth investigating

A standing DISCOVERY Problem asks:

Find objects whose provenance records contain important unresolved gaps or contradictions.

Agents can inspect:

A result may identify:

Object X

Known:
Paris, 1936

Next confirmed:
New York, 1942

Gap:
1936–1942

Candidate leads:
Collector A archive
Dealer B ledger
Auction catalog C

After triage, this can produce a Problem:

Resolve the 1936–1942 ownership history of Object X.

3. Multiple researchers can work independently

Concurrent Attempts are expected.

Attempt A
→ hypothesis: Collector A → Dealer B → Owner C

Attempt B
→ hypothesis: direct sale from Collector A to Owner C

Attempt C
→ newly found archive record contradicting both

Each Submission preserves:

The platform should not force premature consensus.

4. Structured evidence

A provenance Artifact might represent custody events such as:

events:
  - holder: Collector A
    from: 1936
    to: 1938
    confidence: 0.95
    evidence:
      - source: archive_123
        locator: folio 18

  - holder: Dealer B
    from: 1938
    to: 1940
    confidence: 0.62
    evidence:
      - source: auction_catalog_77
        locator: lot 42

The schema can require:

Cheap sandbox checks can enforce those structural rules.

5. Review requires judgment

The ReviewProtocol may ask agent reviewers first:

Does each asserted custody event actually follow from the cited evidence?

Are people and organizations correctly identified?

Does the chronology contain impossible overlaps?

Has conflicting evidence been represented rather than omitted?

Are uncertain inferences labeled as uncertain?

A curator may adjudicate cases where source interpretation is ambiguous.

agent review panel
    ↓
deterministic reduction
    ↓ disagreement
curator

Reviewer reports themselves become useful scholarly evidence.

6. Research generates more research

A successful Attempt may not close the investigation.

It may reveal:

The 1938 transaction appears to involve Dealer B,
but confirming it requires checking Dealer B's private ledger.

That can generate a ProblemProposal:

Search Dealer B's 1938 ledger for Object X.

Another result may suggest:

Determine whether Collector A and Person D are the same individual under different transliterations.

Another:

Investigate ten other objects that passed through the same dealer during the same period.

This is where recursive Problem creation matters.

flowchart TD
    P1[Resolve ownership gap] --> A[Attempt]
    A --> R[Verified finding]
    R --> P2[Search dealer archive]
    R --> P3[Resolve identity]
    R --> P4[Investigate related objects]

7. A provenance record can improve indefinitely

A standing Problem may be:

Maintain the best-supported provenance record for Object X.

Anybody can start a new Attempt when they find:

No Problem Revision change is necessary unless the research standard changes.

Artifacts stay immutable.

The current UI can eventually present a trusted lineage:

record v8
└── based on record v7
    ├── adds Dealer B ledger
    ├── corrects Collector A identity
    └── narrows 1938 transfer date

8. Campaign-wide analysis

Once provenance records follow a shared schema, the consortium can ask:

Which dealers appear most frequently in unresolved gaps?

Which archives would resolve the largest number of open Problems?

Which years and regions contain the densest missing provenance?

What percentage of asserted custody events have primary-source evidence?

Which research strategies most often lead to verified improvements?

This supports operational decisions:

Digitizing Archive Z may unblock 430 open provenance Problems.

The analytical dataset is a derived view over trusted Artifacts, not a new source of truth.

9. What this Campaign requires from TfP

Existing

Stage-8 candidates

Infrastructure

Explicitly not required