Dataset Viewer
Auto-converted to Parquet Duplicate
chunk_id
stringlengths
8
31
record_id
stringlengths
6
29
collection
stringclasses
2 values
title
stringlengths
29
221
url
stringlengths
38
51
year
int32
1.77k
1.81k
⌀
place
stringclasses
66 values
institution
stringclasses
2 values
doc_type
stringclasses
2 values
pages
stringlengths
1
137
text
stringlengths
1
2k
prose
int32
0
2k
ocr_file
stringlengths
10
27
⌀
ocr_run
stringclasses
2 values
blocks
stringlengths
29
6.95k
⌀
image_id
stringclasses
0 values
text_source
stringclasses
2 values
vector
list
sn82014385_1809-07-12_ed1_p1#0
sn82014385_1809-07-12_ed1_p1
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p1
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
1
THE DELAWARE GAZETTE. VOL. I.] WILMINGTON, WEDNESDAY, JULY 12, 1809. [NO. 2.] Printed and Published On Wednesdays and Saturdays, BY JOSEPH JONES, In Market street, a few doors above the Bank of Delaware. CONDITIONS. I. THE DELAWARE GAZETTE shall be published every Wednesday and Saturday, on a large folio sheet. II. T...
1,410
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[0, "/page/0/SectionHeader/0", 0, 21], [0, "/page/0/Text/1", 0, 8], [0, "/page/0/Text/2", 0, 37], [0, "/page/0/Text/3", 0, 8], [0, "/page/0/Text/4", 0, 125], [0, "/page/0/SectionHeader/5", 0, 11], [0, "/page/0/ListGroup/6", 0, 1190], [0, "/page/0/SectionHeader/7", 0, 43], [0, "/page/0/Text/8", 0, 11], [0, "/page/0/Tex...
null
datalab_blocks
[ 0.0156402587890625, 0.01328277587890625, -0.19775390625, -0.07037353515625, 0.007022857666015625, -0.0210723876953125, 0.019287109375, 0.0227508544921875, 0.07611083984375, -0.03375244140625, 0.037353515625, -0.0350341796875, 0.061767578125, -0.00539398193359375, 0.0286865234375, -0.01...
sn82014385_1809-07-12_ed1_p1#1
sn82014385_1809-07-12_ed1_p1
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p1
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
1
Next above the Bank of Delaware, Market Street, Wilmington, Delaware. Each article has on the outside wrapper the signature of Richard Lee & Son. Persons not attending to the above instructions, are liable to be imposed upon. The proprietors think it necessary to remind the public, that these medicines have been for se...
1,570
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[0, "/page/0/Text/11", 0, 126], [0, "/page/0/Text/12", 0, 18], [0, "/page/0/Text/13", 0, 79], [0, "/page/0/Text/14", 0, 392], [0, "/page/0/SectionHeader/15", 0, 5], [0, "/page/0/SectionHeader/16", 0, 25], [0, "/page/0/Text/17", 0, 331], [0, "/page/0/Text/18", 0, 199], [0, "/page/0/SectionHeader/19", 0, 8], [0, "/page/...
null
datalab_blocks
[ 0.0219573974609375, 0.033111572265625, -0.1949462890625, -0.08734130859375, 0.0174560546875, -0.0491943359375, -0.0330810546875, 0.00797271728515625, 0.04193115234375, 0.001094818115234375, -0.0009388923645019531, 0.033111572265625, 0.06292724609375, -0.0143890380859375, -0.0260925292968...
sn82014385_1809-07-12_ed1_p1#2
sn82014385_1809-07-12_ed1_p1
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p1
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
1
For the prevention and cure of bilious and malignant fevers is recommended. Lee's Anti-Bilious Pills. Prepared by Richard Lee and Son Baltimore. Persons wishing to purchase this valuable medicine are requested to be particular in enquiring for Lee's Anti-Bilious Pills, put up in wooden boxes, having on the outside wrap...
1,470
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[0, "/page/0/Text/21", 0, 75], [0, "/page/0/SectionHeader/22", 0, 25], [0, "/page/0/Text/23", 0, 42], [0, "/page/0/Text/24", 0, 274], [0, "/page/0/Text/25", 0, 124], [0, "/page/0/Text/26", 0, 177], [0, "/page/0/Text/27", 0, 121], [0, "/page/0/Text/28", 0, 93], [0, "/page/0/Text/29", 0, 159], [0, "/page/0/Text/30", 0, ...
null
datalab_blocks
[ 0.0306396484375, 0.0239410400390625, -0.1688232421875, -0.083740234375, 0.0309295654296875, -0.0467529296875, -0.04742431640625, 0.005161285400390625, 0.03759765625, -0.013458251953125, 0.00555419921875, 0.038238525390625, 0.08795166015625, -0.01800537109375, -0.0010232925415039062, -0...
sn82014385_1809-07-12_ed1_p1#3
sn82014385_1809-07-12_ed1_p1
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p1
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
1
The Anodyne Elixir, For the cure of every kind of Head-Ache Indian Vedegetable Specific, For the cure of Veneral Complaints. Lee's Laxative Pills. Each and every Medicine above enumerated have got their directions describing their mode of use in the most perfect manner. October 15. James Gardner, RESPECTFULLY informs h...
1,264
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[0, "/page/0/SectionHeader/40", 0, 19], [0, "/page/0/Text/41", 0, 39], [0, "/page/0/SectionHeader/42", 0, 28], [0, "/page/0/Text/43", 0, 35], [0, "/page/0/SectionHeader/44", 0, 21], [0, "/page/0/Text/45", 0, 135], [0, "/page/0/SectionHeader/46", 0, 14], [0, "/page/0/Text/47", 0, 257], [0, "/page/0/ListGroup/48", 0, 92...
null
datalab_blocks
[ 0.004474639892578125, 0.04595947265625, -0.19970703125, -0.0914306640625, 0.010772705078125, -0.02001953125, -0.03509521484375, -0.006317138671875, 0.057586669921875, 0.0214080810546875, -0.0005254745483398438, 0.01294708251953125, 0.1063232421875, -0.025726318359375, 0.033782958984375, ...
sn82014385_1809-07-12_ed1_p1#4
sn82014385_1809-07-12_ed1_p1
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p1
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
1
N. B. Country Store Keepers supplied on the lowest terms for cash or the usual credit. James Gardner. Wilmington, March 25, 1809. tf Notice. SUCH persons as are indebted to the estate of ESTHER YARNALL, deceased, either by bond, note, or book-debt, are requested to make immediate payment to John Warner and John Torbert...
1,372
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[0, "/page/0/Text/49", 0, 132], [0, "/page/0/SectionHeader/50", 0, 7], [0, "/page/0/Text/51", 0, 379], [0, "/page/0/SectionHeader/52", 0, 7], [0, "/page/0/Text/53", 0, 245], [0, "/page/0/Text/54", 0, 67], [0, "/page/0/SectionHeader/55", 0, 24], [0, "/page/0/Text/56", 0, 744], [0, "/page/0/Text/57", 0, 15], [0, "/page/...
null
datalab_blocks
[ 0.031524658203125, 0.031524658203125, -0.2281494140625, -0.047210693359375, 0.00848388671875, -0.006267547607421875, -0.003692626953125, 0.0058746337890625, 0.06060791015625, 0.007778167724609375, -0.035369873046875, -0.031646728515625, 0.102294921875, -0.01181793212890625, -0.0034389495...
sn82014385_1809-07-12_ed1_p1#5
sn82014385_1809-07-12_ed1_p1
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p1
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
1
(BY AUTHORITY.) Scheme of a Lottery, For raising a sum of Money For discharging the Debts due from the Wilmington College, And for repairing the said College. 1 Prize of 5000 dollars is 5000 1 ..... 3000 ..... 3000 3 ..... 500 ..... 1500 1 ..... 400 ..... 400 1 ..... 350 ..... 350 1 ..... 300 ..... 300 1 ...
1,200
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[0, "/page/0/Text/57", 0, 15], [0, "/page/0/SectionHeader/58", 0, 142], [0, "/page/0/Table/59", 0, 377], [0, "/page/0/Text/60", 0, 186], [0, "/page/0/Table/61", 0, 362], [0, "/page/0/Text/62", 0, 236], [0, "/page/0/Text/63", 0, 49], [0, "/page/0/ListGroup/64", 0, 134], [0, "/page/0/Text/65", 0, 131], [0, "/page/0/Text...
null
datalab_blocks
[ 0.004802703857421875, 0.08099365234375, -0.189697265625, -0.051666259765625, 0.02276611328125, -0.037109375, 0.01227569580078125, 0.01338958740234375, 0.06243896484375, 0.0281829833984375, -0.034393310546875, -0.031890869140625, 0.06768798828125, -0.015838623046875, 0.018768310546875, ...
sn82014385_1809-07-12_ed1_p1#6
sn82014385_1809-07-12_ed1_p1
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p1
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
1
Tickets may be had of the trustees, and such persons as they may appoint. February 4, tf NOTICE. THAT the subscriber hath obtained from the orphan's court of Cecil county in Maryland, letters of administration on the personal estate of Abner Kirk, late of Cecil county, deceased. All persons having claims against the sa...
1,437
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[0, "/page/0/Text/67", 0, 88], [0, "/page/0/SectionHeader/68", 0, 7], [0, "/page/0/Text/69", 0, 516], [0, "/page/0/SectionHeader/70", 0, 16], [0, "/page/0/Text/71", 0, 199], [0, "/page/0/SectionHeader/72", 0, 8], [0, "/page/0/Text/73", 0, 89], [0, "/page/0/Text/74", 0, 153], [0, "/page/0/Text/75", 0, 175], [0, "/page/...
null
datalab_blocks
[ 0.007213592529296875, 0.0279083251953125, -0.2005615234375, -0.09832763671875, -0.004451751708984375, -0.0246734619140625, -0.01374053955078125, 0.0328369140625, 0.08966064453125, 0.03790283203125, -0.0120086669921875, -0.02252197265625, 0.06097412109375, 0.01194000244140625, 0.006481170...
sn82014385_1809-07-12_ed1_p1#7
sn82014385_1809-07-12_ed1_p1
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p1
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
1
The American Nepsos a collection of the most remarkable and the most eminent men who have contributed to the discovery, the settlement and the independence of America. Price $1 The Ladies Friend, being a treatise on the virtues and qualifications which are the brightest ornaments of the fair sex. Price 37 1/2 cents. Eu...
1,408
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[0, "/page/0/Text/79", 0, 176], [0, "/page/0/Text/80", 0, 140], [0, "/page/0/Text/81", 0, 97], [0, "/page/0/Text/82", 0, 134], [0, "/page/0/Text/83", 0, 115], [0, "/page/0/Text/84", 0, 90], [0, "/page/0/ListGroup/85", 0, 387], [0, "/page/0/Text/86", 0, 243], [0, "/page/0/SectionHeader/87", 0, 16], [0, "/page/0/Text/88...
null
datalab_blocks
[ -0.0036411285400390625, 0.088623046875, -0.1932373046875, -0.08380126953125, 0.04461669921875, -0.032440185546875, -0.048004150390625, -0.03515625, 0.059967041015625, 0.052398681640625, -0.03253173828125, -0.021728515625, 0.0843505859375, 0.041778564453125, 0.002071380615234375, -0.037...
sn82014385_1809-07-12_ed1_p1#8
sn82014385_1809-07-12_ed1_p1
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p1
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
1
Quakers, methodists, new methodists, jumphers, universalists, rellian universalists, destructionists, sabbarians, moravians, sandemanians, hutchinsonians, dunkers, shakers, new American sect, mystics, swedenborgians, millenarians, together with reflections on the whole, and a recapitulary table. Price $1 To the Public....
1,217
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[0, "/page/0/Text/89", 0, 305], [0, "/page/0/SectionHeader/90", 0, 14], [0, "/page/0/Text/91", 0, 468], [0, "/page/0/Text/92", 0, 18], [0, "/page/0/Text/93", 0, 109], [0, "/page/0/SectionHeader/94", 0, 12], [0, "/page/0/Text/95", 0, 264], [0, "/page/0/SectionHeader/96", 0, 7], [0, "/page/0/Text/97", 0, 279], [0, "/pag...
null
datalab_blocks
[ -0.0333251953125, 0.048980712890625, -0.240234375, -0.0621337890625, 0.0204925537109375, -0.022308349609375, -0.005710601806640625, -0.020263671875, 0.1109619140625, 0.019927978515625, -0.009429931640625, -0.035552978515625, 0.033660888671875, -0.0158233642578125, -0.018280029296875, -...
sn82014385_1809-07-12_ed1_p2#0
sn82014385_1809-07-12_ed1_p2
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p2
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
2
Laws of the Union. [By Authority.] AN ACT To fix the time for the next meeting of Congress. BE it enacted by the Senate and House of Representatives of the United States of America, in Congress assembled, That after the adjournment of the present session, the next meeting of Congress shall be on the fourth Monday of No...
1,563
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[1, "/page/1/SectionHeader/0", 0, 18], [1, "/page/1/Text/1", 0, 15], [1, "/page/1/SectionHeader/2", 0, 6], [1, "/page/1/Text/3", 0, 49], [1, "/page/1/Text/4", 0, 240], [1, "/page/1/Text/5", 0, 134], [1, "/page/1/Text/6", 0, 14], [1, "/page/1/Text/7", 0, 24], [1, "/page/1/SectionHeader/8", 0, 6], [1, "/page/1/Text/9", ...
null
datalab_blocks
[ 0.034149169921875, 0.06170654296875, -0.169921875, -0.044525146484375, 0.06683349609375, 0.00991058349609375, -0.004703521728515625, -0.0025653839111328125, 0.09295654296875, -0.01129913330078125, -0.038543701171875, 0.055816650390625, 0.0227203369140625, 0.006320953369140625, 0.05688476...
sn82014385_1809-07-12_ed1_p2#1
sn82014385_1809-07-12_ed1_p2
chronam
The Delaware gazette (Wilmington [Del.]), 1809-07-12, ed1 p2
https://www.loc.gov/item/sn82014385/1809-07-12/ed-1
1,809
Wilmington [Del.]
LoC
newspaper page
2
J. B. VARNUM, Speaker of the House of Representatives. ANDREW GREGG, President of the Senate pro tempore. June 28, 1809. APPROVED. JAMES MADISON. AN ACT Authorising the discharge of Joseph Wilkinson, junr. from his imprisonment. BE it enacted by the Senate and House of Representatives of the United States of America, i...
1,553
sn82014385_1809-07-12_1.pdf
VLM_OCR_1
[[1, "/page/1/Text/17", 0, 105], [1, "/page/1/Text/18", 0, 14], [1, "/page/1/Text/19", 0, 24], [1, "/page/1/SectionHeader/20", 0, 6], [1, "/page/1/Text/21", 0, 75], [1, "/page/1/Text/22", 0, 795], [1, "/page/1/Text/23", 0, 105], [1, "/page/1/Text/24", 0, 14], [1, "/page/1/Text/25", 0, 24], [1, "/page/1/SectionHeader/26...
null
datalab_blocks
[ 0.052276611328125, 0.052703857421875, -0.157470703125, -0.07513427734375, 0.0555419921875, -0.00769805908203125, -0.0181121826171875, 0.03851318359375, 0.08477783203125, 0.0302276611328125, -0.05145263671875, 0.0293426513671875, 0.047515869140625, 0.0224609375, 0.016571044921875, -0.07...
End of preview. Expand in Data Studio

Dataset Card for Revolution Crossroads Full-Record Search Index

Dataset Summary

This dataset contains 2,106,311 searchable text passages from 169,183 source records brought together for Revolution Crossroads: National Archives Revolutionary War pension files, Library of Congress Chronicling America newspapers, and Smithsonian Institution collection records and object images.

Each passage includes text, source information, and an embedding: a numerical representation generated by an AI model to support searching for related content. The dataset provides a foundation for tools that combine keyword search with search by similarity in meaning. It includes browsable Parquet tables and a packaged search database.

The dataset was prepared as part of the Revolution Crossroads project: https://www.si.edu/revolution-crossroads

Dataset Description

For Revolution Crossroads, the Smithsonian Institution prepared this index to support discovery across three source collections. The text includes AI-assisted OCR, Smithsonian catalog fields, and supplementary pension text where newer OCR was unavailable. It is divided into passages while retaining record identifiers and available page and block references.

Note: This dataset includes AI-generated outputs and may contain errors or omissions. It supports exploration alongside the original records and images. See Quality Considerations and Risks and Limitations for details. Questions and reports of errors can be shared through the Community tab.

Dataset Details

  • Prepared by: Smithsonian Institution
  • Shared by: Revolution Crossroads
  • Language: Primarily English
  • Embedding model: nomic-ai/nomic-embed-text-v1.5
  • Embedding dimensions and storage: 768 values per passage, stored as float16
Collection Records in this Dataset:
Collection Number of Searchable Records Number of Passages Represented
NARA Revolutionary War Pension Files 78,723 1,167,900
Library of Congress Chronicling America 77,793 922,244
Smithsonian Revolutionary Era Collections 12,667 16,167
Total 169,183 2,106,311

A record is a pension file, a newspaper page, or a Smithsonian catalog object. Passages overlap, so their counts do not represent distinct documents or non-repeating text. These counts describe the Index v2 data documented in the repository's manifest and statistics.

Relationship to Source Datasets

The index draws on these Revolution Crossroads datasets:

Some passages retain earlier indexed text where newer OCR was unavailable. The text_source field distinguishes these from passages built from Datalab blocks or catalog fields. The manifest records the source dataset revisions used in the build.

For collection scope, original records, OCR methods, provenance, and rights, refer to the linked source dataset cards. This card focuses on how the text was organized and embedded for search. Complete OCR artifacts are available in the Revolution Crossroads VLM OCR bucket.

The index is separate from the Named Entity Occurrences dataset. Available block references can support connections between search passages and entity mentions when the OCR versions and source identifiers align.

Curation Rationale

The collections differ in format, scale, and description. A shared passage index makes their text available to search tools in a common structure. Keyword indexing supports searches for words, while embeddings support finding passages with related meaning. Retained source information helps users return to the records behind a result.

Dataset Creation

Text was assembled from the sources described above. Consecutive Datalab blocks were grouped into passages of approximately 2,000 characters, with up to 300 characters of whole-block overlap. Longer blocks were split at sentence boundaries. Pension passages may cross pages within a file; newspaper passages stay within one page, and Smithsonian image passages within one image. Smithsonian catalog text retains field labels such as Sitter: and Place made:.

Newly built passages were embedded using nomic-ai/nomic-embed-text-v1.5, with the prefix search_document: added for embedding. Earlier pension and newspaper index rows were retained where no newer Datalab text was available. Vectors were stored as 16-bit floating-point values to reduce size, and keyword and approximate vector indexes were built in LanceDB.

This processing organizes existing text for retrieval. It does not correct the OCR, verify statements in the records, or establish historical relationships.

Data Collection and Processing

The current index contains four text sources:

Dataset Text Sources:
Source Name Number of Passages Source Description
datalab_blocks 2,067,571 Passages built from Chandra 2 OCR blocks.
legacy_fallback 21,277 Earlier indexed text retained for pension and newspaper records without newer OCR text, plus fallback text for two Smithsonian images.
si_fields 13,029 Smithsonian catalog text with field labels.
nara_page_text 4,434 NARA human transcriptions or older extracted text used where new OCR returned no text for a pension page.

Supporting Files Available

Additional Files:
File Path Contents
data/*.parquet Passage table, including text, metadata, source references, and vectors. Organized in subsets by collection for browsing and analysis.
index.tar Packaged LanceDB database containing the chunks table and its search indexes.
manifest.json Model, passage settings, source revisions, vector precision, and retrieval checks.
stats.json Record and passage counts, metadata year range, and document types.

The Parquet files provide access to the table without opening the packaged LanceDB database. The prebuilt keyword and vector indexes are supplied in index.tar.

Accessing a Specific Data Version

This repository provides all, pension, chronam, and si configurations. For example, to stream Smithsonian passages:

from datasets import load_dataset

passages = load_dataset(
    "RevolutionCrossroads/revx-fullrecord-index",
    "si",
    split="train",
    revision="main",
    streaming=True,
)

first_passage = next(iter(passages))

main follows the latest repository contents. For reproducible work, replace it with a specific commit or an available release tag. The manifest describes the index build; release tags documented for the original index repository should not be assumed to exist in this repository.

Quality Considerations

The index inherits errors and omissions in its source text, including AI-assisted OCR. An embedding is a model-generated representation of that text, not a measure of its accuracy. Similarity between passages does not establish a historical connection. The source cards provide further information about OCR quality and collection-specific limitations.

The build included internal checks of identifiers, text, vectors, and retrieval behavior. At the documented search settings, approximate vector retrieval returned an average of 94.8 percent of the top ten results from exact float32 search over 25 test queries. This measures agreement between search methods on that test set; it does not measure OCR accuracy or whether the results answer a historical question.

Dataset Structure

Each row represents one passage, identified by chunk_id. Several passages can share a record_id, and overlapping passages may repeat text. all contains every collection; the other configurations contain one collection each. The split name train is a storage convention and does not identify a manually reviewed training set.

Passages retain different source references according to text_source. Datalab-derived passages have OCR file and block references; catalog and fallback passages may lack them. The pages field records source-facing page numbers as text, while page indexes inside blocks are zero-based. These conventions should be accounted for when matching passages to source images or OCR blocks.

Data Fields

The Parquet table contains 18 fields. Types below match the published viewer schema. Values may be null or empty where information is unavailable.

  • chunk_id (string)
    Unique passage identifier.
    Example: 111404018#0

  • record_id (string)
    Identifier of the searchable record: a pension NAID, a newspaper page identifier, or a Smithsonian EDAN identifier.
    Examples: 111404018, sn82014385_1809-07-12_ed1_p1, edanmdm:nmah_1000572

  • collection (string)
    Collection label: pension, chronam, or si.

  • title (string)
    Record title supplied for display and context.

  • url (string)
    Link to the corresponding source record or newspaper issue.

  • year (int32)
    Record-level year used for filtering, where available. It does not describe every date mentioned in a passage.

  • place (string)
    Record-level place information used for display and filtering.

  • institution (string)
    Source institution or Smithsonian museum associated with the record.

  • doc_type (string)
    Document or object type used for display and filtering.
    Examples: pension file, newspaper page, scutching knife

  • pages (string)
    Source-facing page numbers represented in the passage, stored as text rather than a list. Empty for Smithsonian passages.
    Example: 1,2,3,4

  • text (string)
    Passage text used for keyword search and embedding.

  • prose (int32)
    Count of letters in words used to assess the amount of prose. This is not an OCR accuracy score.

  • ocr_file (string)
    Datalab processing filename, where available.
    Example: 111404018.pdf

  • ocr_run (string)
    OCR run associated with the passage, where available: VLM_OCR_1 or VLM_OCR_2.

  • blocks (string)
    JSON-encoded list of source block references in the form [page, block_id, start, end]. The page index is zero-based; the character bounds identify the portion of the source block used. This field is stored as a string, not a nested list column.
    Example: [[0, "/page/0/Text/1", 0, 7]]

  • image_id (string)
    Smithsonian image identifier for an image-derived passage, where available.

  • text_source (string)
    Passage origin: datalab_blocks, legacy_fallback, si_fields, or nara_page_text, as described above.

  • vector (fixed-size list of float16, length 768)
    Passage embedding generated with nomic-ai/nomic-embed-text-v1.5. Query embeddings for compatible vector search use the same model with the prefix search_query: .

Considerations for Using the Data

Risks and Limitations

  • Search coverage is incomplete. Source material without usable text is not searchable. Some near-empty pension pages were excluded, and a represented record need not have text for every page.
  • Text sources vary. Newer OCR, catalog text, transcriptions, and earlier fallback text have different characteristics. text_source helps identify those differences.
  • Passages overlap. Counting passage rows or matching text across them can overcount documents, words, or mentions.
  • Retrieval is approximate. Similarity rankings and finite candidate pools can omit relevant passages. The packaged keyword index stores no word positions; applications that implement phrase matching over a candidate pool may miss matches outside it.
  • Metadata filters have limits. Years, places, and document types may be missing or inconsistent and should not be treated as a complete description of passage content.
  • Historical language is retained. Source text may contain outdated or offensive terminology. OCR can also introduce incorrect names or statements.

Recommendations

Use record_id to distinguish records from passages and retain source references when reusing text. For retrieval experiments, document the dataset revision, model, and search settings. Inspect promising passages in their original records and images before relying on a reading or interpreting a connection.

Additional Information

Licensing Information

For rights and reuse information about the underlying records, images, and extracted text, consult the linked source dataset cards and originating institutions. The embedding model and database software have their own license terms, available through their documentation. Consult the embedding model card for model-specific information.

Citation Information

BibTeX

@dataset{RevolutionCrossroads_Fullrecord_Index_2026,
  author = {Revolution Crossroads Project Team},
  title = {Revolution Crossroads Full-Record Search Index},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/RevolutionCrossroads/revx-fullrecord-index}
}

APA

Revolution Crossroads Project Team. (2026). Revolution Crossroads Full-Record Search Index [Data set]. Hugging Face. https://huggingface.co/datasets/RevolutionCrossroads/revx-fullrecord-index

Glossary

  • Passage or chunk : A portion of a record's text represented by one row. Neighboring passages may overlap.
  • Embedding or vector : A numerical representation of text generated by a model for comparing similarity.
  • Hybrid search : Retrieval that combines keyword and vector search rankings.
  • Index : A data structure prepared to make searching faster.
  • Fallback text : Alternative or earlier text used where newer Datalab text was unavailable.

Dataset Card Contact

revolutioncrossroads@si.edu

Downloads last month
23