Purpose
`pnpm research:novelty` evaluates a structured cross-source historical lead. It
is designed to prioritize archival research, not manufacture novelty claims.
The default input is the unresolved Rosenberg/RBS *5036 case.
What it measures
identifiers.
probably matching object as still missing.
more than one retained prior-art search.
- Object-match signals: normalized artist, title tokens, dimensions, actor, and
- Dimension relationship: exact, reversed, different, or unknown.
- Chronology: whether a reported sale predates a source that describes the
- Research completeness: decisive identity evidence, transaction evidence, and
Scores never verify identity. Even the strongest score is
`probable-not-proven`; only inspected object-level documents can bridge records.
Every generated report remains publication-ineligible.
Current lead
RBS Tome II entry 5036 lists a Georges Braque, Des fruits et une guitare*,
81 × 100 cm, under Paul Rosenberg's OBIP 37.954 claim. The catalogue's dated
preface says already recovered works were excluded. Christie's later published
provenance for an 81 × 100 cm 1938 Braque, *Nature morte à la guitare (rideaux
rouge)*, says Rosenberg sold it to the Laskers in January 1947.
The initial workbench score was 85/99. Deeper research found a second 1938
Braque with the same 81 × 100 cm dimensions: the National Gallery of Art's
Fruit, Glass, and Mandolin, cited to Mangin page 33, adjacent to the
Christie's/Lasker work on page 34. With the known-collision penalty, the score is
now 60/99 (`possible`) and no chronology conflict qualifies without an
object-level bridge. Novelty remains `unverified`.
Decisive archive targets
The MoMA Paul Rosenberg Archives finding aid narrows the next review to:
receipts spanning 1946–1950.
- `III.A.2.2.42`: Braque, “Fruits and Guitar,” 1938.
- `III.D.1.c` and `III.D.1.h`: Braque inventory/index cards.
- `V.A.2` and `V.A.12`: missing-picture lists and photographs, 1940–1947.
- `II.QQ.1` and `II.QQ.8`: detailed painting lists, sales, invoices, and
- OBIP claim file 37.954 and Mangin's 1961 catalogue, page 34.
The ready-to-send request is retained in
`docs/product/rosenberg-archive-reproduction-request.md`. MoMA's form is closed
for annual maintenance until September 8, 2026.
An official French restitution workbook adds a second bounded lead. Row 1110
indexes a Braque, Grande nature morte, number `15 ?`, returned to Paul
Rosenberg on September 14, 1945. POP record `JDP-0056` uses the same generic
title and a Jeu de Paume room number 15 but reports restitution on September
12, 1947. This is not yet a contradiction: the number systems may differ and
the records may concern different paintings. The retained receipt is
`artifacts/provenance-research/rosenberg-restitution-discharge-receipt-v1.json`.
The archive inventory now locates the correct primary target in `209SUP/1`,
dossier `45.15 - M. Paul Rosenberg`, views 199–1173. Its restitution section
separately lists a September 14, 1945 judicial-method return and a September
12, 1947 return of works found in Belgium. The two dates therefore do not by
themselves form a contradiction. The decisive question is whether the Braque
appears in one primary list or both.
The earlier `209SUP/392` target is rejected. It is an E–I volume, and its viewer
media 33–38 are unrelated Hirschberg-Bellet material. The workbook's references
to pages 33–38 and 45–48 are internal document pagination, not viewer media
indices. This negative result prevents a false primary-document bridge.
The correct dossier has now supplied a positive bridge for the separate 1947
Braque. Internal page 33 (viewer media 1154) describes a Braque Nature Morte,
97 × 130 cm, found with a Brussels picture dealer and claimed by Paul
Rosenberg. Internal page 48 (viewer media 1173) calls the work *Mandoline,
fruits et bouteille*, reports recovery in Brussels, and says it was to be
restituted to Rosenberg after formalities. Image URLs, byte counts, hashes,
supported transcription, and limitations are retained in
`rosenberg-belgium-braque-primary-bridge-v1.json`. These letters support the
recovery process and intended restitution, not the final September handover.
Interpretation boundary
The present evidence does not show an unrestituted painting, wrongdoing, false
title, or inaccurate provenance. Plausible explanations include catalogue lag,
an imprecise sale date, an erroneous claim entry, or two distinct works. A
bounded negative web search is not proof that no scholar has published the
discrepancy.
Reproduction
pnpm research:novelty
node --import tsx --test tests/services/provenance-novelty-workbench.test.ts
Inputs and reports live under `artifacts/provenance-research/`. Add evidence by
editing a versioned input; do not overwrite an earlier research claim silently.