An independent educational project

From three handwritten cells to a structured record

A bounded, AI-assisted reading of an approved excerpt from the St. Croix General Tabell of 31 December 1792, with provenance and limitations.

A permitted excerpt from existing transcription work: three cells, not a full dataset or a validated linkage.

1. Locate the historical source

Rigsarkivet: St. Croix General Tabell, 31 December 1792. Dronningens Quarteer, third estate row; viewer image 207. Identifiers: epid 20201795 · bsid 292226 · imageID 58788329.

Crop of the historical column headings, including Creoler and Africaner.
Column headings. Historical labels are not categories endorsed by this project.
Crop of three handwritten counts, read as 54, 32 and 86.
Three cells from the same row. The crop contains no personal names.

Source: Rigsarkivet. Two rectangular crops of the same scan; the numerals have not been altered. Permission for public reproduction confirmed by Anton on 24 October 2026; coauthor approval also confirmed. No general reuse licence is granted.

2. Preserve a bounded structured record

Unit: persons recorded as enslaved. The historical origin categories are retained as administrative labels, not validated personal identities.

Three values from one row, 31 December 1792
FieldSource labelValue
creolesCreoler54
africansAfricaner32
totalTotal86

Download the three values (CSV) · Download provenance and checks (JSON)

3. State exactly what was checked

What the checks do not establish

Misread numbers can still add up correctly. This example is not independent human ground truth, a full-sheet accuracy test, a genealogical link or a thesis finding. No linkage across years is demonstrated here.

4. Make the trail inspectable

The provenance file records the archival link, identifiers, image dimensions, crop coordinates and SHA-256 hashes of the scan and reading log used. These identify inputs; they do not make the reading infallible. The full scan and private reading log are not released here.

A separate methods note: number evidence is not name evidence →