How a Saint Entry Is Made

Two Halves, Made Differently

Saints Oil Paintings gathers pictures of saints. It does not identify them. Every entry is welded from two things that share nothing but a page: a catalogue block — painter, date, support, measurements, the collection holding the panel — copied out of public collection data, and prose underneath it drafted by a language model reading those same fields.

We print the seam instead of hiding it, because the halves earn different amounts of trust. A measurement taken from a museum's own database can be leaned on. A paragraph on what Catherine's broken wheel meant to a fifteenth-century viewer is orientation, and deserves the scepticism you would give an unsigned wall label.

Where Each Field Comes From

The catalogue block

Open museum APIs, Wikidata and the Getty Union List of Artist Names. Where a value has been checked against one of those, the entry prints that source beside it as a numbered reference. Where it has not, the value stands exactly as it arrived and nothing is implied — which today describes most of the collection. An entry showing no Sources list has not been checked against one.

The prose

Drafted by a language model from the fields above. It is not hagiography, not archival scholarship, and not the work of a named historian of the cult of saints. Follow the references before quoting any of it.

The plate

Public-domain reproductions with the holding institution's attribution kept intact. Terms are on the copyright notice.

What the Attribute Cannot Do Here

A saint is recognised in Western painting by the thing standing next to him: Catherine's wheel, Lawrence's gridiron, Sebastian's arrows, Jerome's lion and his hat. Reading those emblems and naming the figure is hagiographic work, and this site does not do it. We repeat the title the museum gave the picture. Where that title says only "A Female Saint", so do we.

Two failure modes follow, and naming them beats letting a reader trip over them. First, saints share names: Catherine of Alexandria and Catherine of Siena are two women with two sets of attributes, and a bare "St Catherine" in a source record does not say which one is on the panel. Second, half the map of France is named after somebody canonised. The collection is built by matching whole words in the title, with an explicit rule that drops place names — Mont-Saint-Michel, the Ile Saint-Louis, the asylum at Saint-Remy — so a river bend does not arrive dressed as a martyr.

That rule is a filter, not a judgement. A stray village view can still slip past it, and a real saint whose title reads like a street sign can still be kept out. Both errors are ours, and both are worth reporting.

What We Do Not Claim

Other reference sites assert the following. Here they would be false, so we assert none of them:

• No referee has read a word of this. There is no peer review.
• No hagiographer, theologian or curator sits on a board here, because there is no board. Nobody on this site holds a degree in the subject.
• Nobody has been interviewed — not a curator, not a parish, not a scholar.
• No unpublished document has been opened. The Bollandists' Acta Sanctorum and the Golden Legend are where saints are actually argued about; we read neither.
• The museums whose open data we read are not partners of ours. Calling a public API is not a collaboration.

A small operation described accurately is worth more than a large one described falsely. If anything elsewhere on this site contradicts this list, the list wins — tell us and the other page gets corrected.

Human Review

Nobody is published as editor of record. Until somebody takes that name, assume the prose on an entry has been read by no human being and lean on the catalogue block and the sources it lists. Printing a masthead with an empty chair behind it would have been the easier thing to do.

The collection is assembled by Louie Drehmer, who decides which panels enter it and runs the pipeline that fills the fields. Choosing pictures and checking facts are different jobs, and the first is not allowed to stand in for the second here.

When Two Records Disagree

Dimensions taken before a relining, a date printed as "c. 1480" in one catalogue and 1476 in another, a panel that changed hands while half the databases slept: disagreement between records is ordinary, not a defect. We neither settle it nor show you both readings. Collections are queried in a fixed order, the first value found for a field is kept, and everything a later source would have said about that field is discarded rather than stored.

What the entry gives you instead is the source under the value, named and linked, so you can open it and hold it against another catalogue yourself. Smaller than resolving the conflict for you. It is also what the software actually does.

Attribution behaves identically. Values are copied verbatim, so a "workshop of" or an "attributed to" already in the record survives into the entry intact. We never add that hedging ourselves, and its absence proves nothing about how secure an attribution is.

Corrections

Write to editors@saintspaintings.com with the entry URL and what is wrong with it. A catalogue fact gets corrected against the source record, and the entry then carries the day its facts were last checked, printed on the page rather than filed in a changelog nobody opens.

Anything that changes the substance of an entry moves that date. Where a claim cannot be settled either way, the entry says so instead of quietly losing the disagreement.