The Catalogue Changes Its Mind

In 1836 the London dealer John Smith catalogued Rembrandt's paintings; Hofstede de Groot's English volume VI, with a preface signed in April 1915 and a 1916 copyright notice, printed a table saying where each Smith number went. Read cell by cell from the page images, the table keeps 535 of 627 Smith-number rows and inner references and sends 92 elsewhere, 43 of them to named painters. Guess what became of each Old Testament subject, then take the census apart: entries described from a print, etching, engraving, or drawing were dropped more often than those naming a present owner or place, and the two printings of the alignment disagree 8 times.

The Comparative Table keeps 535 of Smith's 627 rows and inner references, sends 92 elsewhere, including 43 to named painters. Hofstede de Groot's English volume VI has a preface signed in April 1915 and a 1916 copyright notice.

Smith 4, and what became of it

John Smith's catalogue gave every Rembrandt he knew of a number. Seventy-nine years later, in a preface signed at The Hague in April 1915, Hofstede de Groot wrote his own catalogue on top of Smith's, and at the back he printed a Comparative Table: for each Smith number, the number the same picture has in the new catalogue, or a footnote saying why it has none. The table is the later catalogue's verdict on the earlier one, printed in a column of numbers.

The strip is Smith's first section, subjects from the Old Testament, numbers 1 to 55. It opens on number 4, Abraham dismissing Hagar. Read Smith's own words, choose what you think Hofstede de Groot did with the picture, then ask for his answer.

Smith 1836, subjects from the Old Testament. Choose a number.

What did Hofstede de Groot do with it?

Smith 4 is the page's opening case because it is the one a comparison of the two texts gets wrong. Smith describes a composition Engraved in mezzotinto by Spilsbury, and described from the print, and then, in a second paragraph, mentions a picture of the subject sold at the Hague. Hofstede de Groot pulls those apart. The engraved composition goes to a footnote, engraved by Spilsbury from a school-piece; the picture from the sale becomes a row of its own, 4*, and an entry of its own, 6a, known only from the sale. The two catalogues do not disagree about one picture. They disagree about how many pictures there were.

The whole catalogue

The table has a cell for all 620 of Smith's numbers, and 7 more rows for inner references to pictures Smith mentions inside an entry. Under the current choices, it keeps 535 of those 627 rows or references as Rembrandt entries or cited mentions. It gives 43 to named painters, among them Ferdinand Bol, Aert de Gelder, Govert Flinck and Salomon Koninck; calls 23 school-pieces and 14 copies; rejects 8 outright or as forgeries; codes 3 as print-only or not established as a picture by Rembrandt; and cannot identify 1.

Kept means catalogued as Rembrandt's. It does not mean the picture could be found: many kept entries are known to Hofstede de Groot only from an old sale or a print, and some he doubts in the entry itself. And a number is not a picture. The table marks 60 of Smith's numbers = Sm. another number, and sends other groups of numbers to one entry, so under the current choices the 627 rows or references form 560 groups after the page's explicit identity unions, 478 of them kept.

By Smith's own sections the losses are uneven. Of his 20 landscapes Hofstede de Groot sends 11 away; of the 56 Old Testament rows, 13. The section table above has the rest under the current choices.

Eighty years of progress?

The obvious objection: Hofstede de Groot simply corrected Smith with eighty more years of evidence, as later scholars corrected him. The differences are progress, nothing more.

Four results say the change is not one thing, and each can be taken apart below.

What Smith had in hand

Smith states how he knew a picture in a few fixed phrases: described from the print, Now in the collection of, a sale line with a price. Coding each of his own entries by the first evidence category it contains, the entries he described from a print, etching, engraving, or drawing were dropped more often than the ones whose present owner or place he could name: 31 of 136 in the print, etching, engraving, or drawing category are not kept, against 31 of 262 naming a present owner or place and 11 of 119 known from a sale record only. If later evidence had simply improved the same verdicts, how Smith knew a picture should not predict what happened to it. It does. What the counts cannot say is why: a picture Smith knew only from a print may simply be harder to find again, and a print can be made after a pupil's picture as easily as after the master's.

What counts as Smith naming where the picture was

a named present owner or place that, or a recent exhibition

The same rule applied to a transcription made from the page images agrees with the OCR coding on 38 of 40 entries drawn at random, and on 55 of 55 Old Testament entries. The two misses are OCR damage: Now iii the collection, and a sale year read as 180Q.

Five kinds of difference

For the Old Testament run the page sets each Smith entry beside the entry the table sends it to and lists every difference: title, subject, date, inscription, size, owner, value, doubt, attribution, identity. Each difference gets exactly one class by a written rule. Of 184 differences, 54 are judgement: the attribution, the identity or the subject of the picture, a date as judged, a doubt. 68 are the world moving on: new owners, cities renamed, inscriptions read on pictures Smith never saw, a different measurement. 21 are Hofstede de Groot repeating Smith, and every one of them agrees with Smith's words once his guineas are turned into pounds. 32 are the same fact written another way. 9 are not in either book at all but in the Archive's machine reading of Smith's pages: Smith 2 prints done in 1646, and the OCR says done in 1648.

Judgement is real, and some of it is striking. Smith's Lot in a Cave becomes JEREMIAH MOURNING OVER THE DESTRUCTION OF JERUSALEM.; his Hannah instructing her Son Samuel becomes Timothy and his Grandmother (or, The Nun and the Child). But it is under a third of what differs.

Old Testament run, Smith 1836 against Hofstede de Groot's English volume VI, one row per difference.

Hofstede de Groot against himself

Hofstede de Groot printed his alignment twice: once in the table, and again after the title of each entry, as in 25. MOSES BREAKING THE TABLES OF THE LAW. Sm. 25. The page checked the heading of every one of the 484 entries the Smith column points to. In 476 the heading cites the Smith number. In 5 it does not, and the heading one entry further on does: the table sends Smith 148, 365, 381, 396 and 409 to entries 248, 249, 250, 251 and 254, while it is the headings of 249, 250, 251, 252 and 255 that cite them. That is what a numbering shifted by one after the table was set would look like; the record does not say that it happened. In 3 more, no heading cites the Smith number at all. The authority choice above decides whose word to take, and the kept count moves from 535 to 532 or 527.

Smith against Smith

Smith revised himself. In his Supplement of 1842 he points back to his own 1836 numbers 19 times, 2 of them to correct a name. Hofstede de Groot's Sm. Suppl. column follows him 16 times and overrules him 3 times, sending the later picture somewhere other than the earlier number Smith named: Supplement 17 (See also, No. 416), 27 (probably the same picture described No. 203) and 28 (See 215). Once, at Supplement 15, he makes an identification Smith never made. Crediting all of those changes to Hofstede de Groot would give him Smith's own second thoughts.

Ask the record

Type a Smith number, a Supplement number or a Hofstede de Groot number. Some questions this record cannot answer, and the page refuses them and says why: a number the table does not print, what anyone thinks now, whether a picture was destroyed, whether an entry with no Smith number was found after Smith wrote.

Question

The check

The record is loading; the figures on this page have not yet been recomputed.

The browser check above reruns the displayed figures, loaded-file hashes and sample. The following are build-time verifier assertions, so this panel names them rather than presenting their static prose as a live browser result:

anchor:supplement-2-7-9-10-11-reproduced-by-engine
route:heading-citations-recomputed-from-readings
fault:ocr-1648-put-back-into-smith-2-raises-the-restatement-alarm
refuse:not-in-table:Sm. 621
provenance:sha256-matches

The anchor, a build-time assertion

No published count of what Hofstede de Groot's table did with Smith's numbers was found, so the census has no outside number to match. The anchors prove instead that the transcription and the reader read the two books as printed. Smith's 1842 pointers for Supplement 2, 7, 9, 10 and 11 (See No. 150, Vol. vii. and the rest) are parsed from Smith's text by the same function that reads all 19, and compared with the table's Sm. Suppl. column read by the same cell reader the census uses: all five agree. Hofstede de Groot's own printed example, Thus, Wurzbach 6 = HdG. 755., is returned by that cell reader from the Wurzbach column of p. 456 as 755.

The transcription, by two routes, checked at build time

The Smith column was read from the full-resolution page images of pp. 456 to 472, 629 rows. The Archive's OCR of the same pages agrees on 537 rows, is unreadable on 75 and differs on 17; each of the 17 was settled on the image, most of them italic letters the OCR reads as stars. The second route is Hofstede de Groot's own headings. Of the 484, 418 cite the Smith number in the OCR exactly as printed, 24 do once OCR confusions are read (such as 8111.617 for Sm. 617), 25 were matched by hand on garbled OCR, and 17 were read on crops of the page image, including every disagreement above. The 82 footnotes the two Smith columns cite were transcribed from the images.

The Old Testament entries of both books, and 40 random Smith entries, were transcribed a second time from the page images by a reader who did not see the OCR, and compared with it; disagreements on sizes and dates were settled on crops.

The rules

Footnote coding, first match wins: not now to be identified is unidentified; not a picture or nothing to show that is coded as print-only or not established as a picture by Rembrandt; copy is a copy; school-piece or by a pupil is a school-piece; replica is a copy; not by Rembrandt, forgery, imitation or had nothing to do with it is rejection; by followed by a painter's name, after engravers and photographers are removed, is another painter; doubtful is doubtful.

Smith's evidence, first match wins: described from a print, etching, engraving or drawing; Described from the sale catalogue or another catalogue; Now in, in the possession of, belonging to, the property of; with the exhibition choice, Exhibited; a sale line; an engraving mentioned; none of these.

Five-way classes: attribution, identity, subject, a date as judged and doubt are judgement; an inscription, an owner, a place and a size Hofstede de Groot reports that Smith does not give, or gives differently beyond the size tolerance, are world; anything Hofstede de Groot attributes to Smith is restatement, checked number by number against Smith's own words, with guineas at twenty-one shillings; a title with a name in common, and a size within the tolerance, are convention; digits and years where the OCR of Smith's page differs from the page are transcription. A subject counts as changed only if no name in Smith's title appears in Hofstede de Groot's. At a two-inch tolerance world falls to 62 and convention rises to 38.

The choices, and what each moves

At the corresponding alternative settings: counting a copy of a catalogued picture as its composition kept gives 546 kept; including Smith's Supplement gives 657 units; folding pictures mentioned inside an entry gives 620; applying Hofstede de Groot's own 4 statements that one of Smith's pictures is possibly or probably identical with another gives 556 groups.

The sample

Provenance

What stays uncertain

No independent published count of these verdicts was found. The census rests on the transcription and on the rules above, both shown.
Kept means catalogued by Hofstede de Groot as a Rembrandt entry. It includes pictures lost since the eighteenth century and pictures he doubts in the entry.
The evidence cross-tab is coded from OCR text. Its stated error is the sample above.
The sample proves agreement between the same rule applied to OCR and an image transcription, not that the phrase rule itself is historically correct.
The Old Testament image comparison uses a partial transcription of title, date, size, provenance and hedge fields; the image transcriptions were made by Claude subagents, not a human.
The five-way classes are this page's rules, not the catalogues'. A reader could class some differences otherwise, especially a retitled subject.
The relaxed OCR heading matches were made on garbled OCR rather than image checks; the page-image route is retained for every table and heading disagreement.
Which printing is right where the table and the headings disagree is not settled by the record, so it is a choice.
Rows such as 4* and 287a both mark a second picture mentioned inside one Smith entry. Why Hofstede de Groot uses a star for some and a letter for others is not stated.
The table prints 662A and 520A with capital letters where the entry headings use a lower-case letter; the page treats them as the same entries.
The English volume is dated 1916 by the Internet Archive's copyright record; its preface is signed at The Hague in April 1915. It translates the German edition published by Paul Neff in Esslingen.

Sources and what came before

John Smith, A Catalogue Raisonné of the Works of the Most Eminent Dutch, Flemish, and French Painters, Part the Seventh (Rembrandt), London: Smith and Son, 1836. Internet Archive.
John Smith, Supplement to the Catalogue Raisonné, Part the Ninth, London, 1842. Internet Archive.
C. Hofstede de Groot, with Karl Lilienfeld, Heinrich Wichmann and Kurt Erasmus, translated by Edward G. Hawke, A Catalogue Raisonné of the Works of the Most Eminent Dutch Painters of the Seventeenth Century Based on the Work of John Smith, vol. VI, London: Macmillan, English witness with a 1916 copyright notice and a preface signed in April 1915, translated from the German Paul Neff edition in Esslingen. Internet Archive.

The alignment itself is not new: Hofstede de Groot's Comparative Table printed it, and the page reads that table rather than guessing at matches. The Wikimedia Commons gallery of his Rembrandt catalogue points readers to that concordance; English Wikipedia lists later Rembrandt catalogues; A Web Catalogue of Rembrandt Paintings compares catalogue totals; the Rembrandt Database's overview of oeuvre catalogues places Smith's volume in the line that runs to the Corpus of Rembrandt Paintings; Wikidata's Hofstede de Groot project aims at an item for every painting he documented, and its summary counts Smith's Rembrandt entries as 620 plus 32, the same as this transcription. Gary Schwartz's Cutting and pasting from 1836 to 2025 (2025) follows a single Smith entry, number 393, into Hofstede de Groot's entry 288; row 393 of the table as read here prints 288, an outside check on one cell. We searched the Artificial Wasteland corpus, Hofstede de Groot's Comparative Table, Wikimedia Commons, English Wikipedia's Rembrandt catalogue lists, the Web Catalogue of Rembrandt Paintings, Wikidata's Hofstede de Groot project page, the Rembrandt Database and open web searches on 2026-09-14, and did not find a published count of what Hofstede de Groot's table did with each of Smith's Rembrandt numbers, nor an interactive that separates changed judgement from changed ownership, restatement and transcription for these two catalogues.

Smith's books are public domain by date. The English volume of Hofstede de Groot is public domain in the United States by date; its status elsewhere is not established, so this page ships its facts and short quotations with page references and links to the Archive's page views, and re-hosts no page image. Terms for every source are in LICENCE.txt beside this page.