Editorial standards

This page describes practice, not aspiration. If we depart from it on a given page, that is a failure worth reporting to lumingrade@gmail.com.

Sourcing

Two kinds of source do most of the work here.

Primary sources are the objects themselves: the 1665 first edition, the plate as printed, the manuscript profile, the instrument. Where a digitized scan exists we read the scan, not a description of it. When we say a plate shows something, someone looked at the plate.

Named institutional repositories supply both the scans and the catalog metadata we rely on for dates, editions, shelfmarks, and provenance. We name the repository every time. "A museum in London" is not a citation; a named institution with a linked catalog record is.

Secondary literature — monographs, journal articles, published catalogs, critical editions — is used for interpretation and for dates that primary material does not settle. It is cited specifically: author, title, year, and a page or section where the claim can be found. We prefer peer-reviewed and archivally published work over encyclopedias, and we do not cite an encyclopedia as the sole support for a substantive claim.

Sources we do not treat as sufficient on their own: undated web pages, auction listings, image aggregators that strip provenance, social posts, AI-generated summaries, and content farms. Any of these may point us toward a real source; none substitutes for one.

Every collection record and every Journal essay ends with a sources list. If a claim in the body is not traceable to something in that list, the claim should not be there.

Fact-checking

Each page goes through the same sequence before it is published:

1. Claim extraction. The draft is read back and every factual assertion is pulled out as a separate line — dates, titles, editions, attributions, measurements, quotations, sequences of events, and any statement about what an image depicts.
2. Source matching. Each line is matched to a specific source and a specific location in it. Lines that cannot be matched are cut or rewritten as an explicit uncertainty.
3. Primary check. Where the claim concerns the object, it is checked against the object's own digitized surrogate or the holding institution's catalog record, not against a secondary retelling.
4. Quotation check. Quotations are transcribed from the source text and compared character by character. Original spelling and punctuation are kept; any modernization is flagged in the caption or a bracketed note. Ellipses mark omissions.
5. Image check. Each image is confirmed to be the item it is captioned as — right edition, right plate number, right figure — and the credit line is confirmed against the source page.
6. Uncertainty pass. Contested dates, disputed attributions, and claims resting on a single source are labeled as such in the text. We would rather write "the surviving evidence does not settle this" than pick a side for readability.

Where a subject sits outside our competence, we ask someone who knows it. Where we cannot get that review, we say less.

How public-domain status is assessed

We publish images only where we can articulate a reason the underlying work is in the public domain, and only where the source's own terms allow reproduction. Assessment is per item, and it is recorded on the record.

What we look at, in order:

• The underlying work. Date of creation, date of first publication, place of publication, author's death date where relevant, and the copyright term that applies in the United States, where this site is published. Works published long enough ago that any term has expired are the clearest case, and most of the collection sits there.
• The digital surrogate. A faithful photographic reproduction of a flat public-domain work is treated as carrying no separate copyright of its own in US practice. We note, plainly, that some institutions and some jurisdictions take a different view, and we respect an institution's stated terms even where we believe the reproduction itself is not separately protected.
• The source's stated terms. Whatever rights statement the holding institution publishes — public domain mark, a Creative Commons designation, "no known copyright restrictions", or a restrictive license — is recorded verbatim on our record and followed. A restrictive statement is respected even when we think the work is free; the institution's relationship with its own material is not ours to override.
• Non-copyright constraints. Some material carries obligations that are not copyright: cultural sensitivity, donor conditions, privacy of identifiable living people, or human remains. These can be reasons not to publish even where copyright has plainly lapsed.

Three practical rules follow. Where status is uncertain, we describe the item in words and link to the source rather than reproduce the image. Where an institution asks us to stop, we stop first and discuss afterward. And where we have made a judgment call, the record says which way we called it, so a reader can disagree with the reasoning rather than guess at it.

None of this is legal advice, and it does not transfer to your use of the same material. If you intend to reuse an item, verify its status for your jurisdiction and purpose. Our terms of use say the same thing at more length.

Image credits

Every reproduced image carries a caption that gives, at minimum: what the image shows; the work it comes from with date and edition; the holding institution; the rights statement the institution applies; and a link to the source page for that image. Where the institution supplies preferred credit wording, we use their wording.

What we do to images: scale them, compress them for the web, and occasionally crop to a single figure where a plate holds several. What we do not do: alter content, composite two items into one picture, colorize, remove foxing or damage, straighten or clean up a plate so it reads as better preserved than it is, or generate any part of an image. A crop is disclosed in the caption. Anything beyond a crop and a resize would be disclosed too, and we have no plans to do it.

We do not use AI-generated or AI-enhanced imagery anywhere in the collection or the Journal. An image on this site is a photograph of a real object made by the institution that holds it.

Corrections policy

We expect to be wrong sometimes, and we would rather find out from a reader than not find out.

Send errors to lumingrade@gmail.com. A useful report names the page, quotes the sentence, and says what the correct fact is — a source for the correction helps, though we will check it either way. Reports are acknowledged and worked through in the order they arrive.

How we handle what comes in:

• Corrections are appended, not hidden. When a substantive error is fixed, a dated correction notice is added to the foot of the page saying what was wrong and what it now says. The page is not silently rewritten. If you read something here and quoted it, you will be able to see that it changed and how.
• Substantive means anything that changes meaning — a date, an attribution, a figure, what an image depicts, a quotation, a causal claim, or the sense of a sentence. All of these get a notice.
• Typographic fixes are not logged. Spelling, punctuation, a broken link repointed to the same target, and markup repairs are corrected without a notice, because logging them would bury the notices that matter.
• Serious errors get more than a footnote. If a claim was central to a piece and it was wrong, the correction goes at the top, not the bottom, and the standfirst is amended to match.
• Withdrawal is visible. If a piece cannot be repaired, we retract it in place — the URL keeps working and carries an explanation of what was wrong and why the piece is gone. We do not delete pages to make an error disappear.
• Disagreement is distinguished from error. Where a reader disputes an interpretation rather than a fact, and the point is substantive, we may note the disagreement in the text instead of changing the conclusion. We will say that is what we have done.

Funding and independence

Nothing on this site is paid placement. We take no sponsored posts, no affiliate commissions, no advertising, and no payment or consideration of any kind for including an item in the collection, for writing about it, or for linking to an institution. We do not sell links. We do not accept copy from a third party and publish it as our own.

How the work is paid for:

If that changes — if we take a grant, a sponsor, or reader funding — this section is updated before the money affects anything published, and any piece connected to a funder carries a disclosure in its byline. Editorial decisions rest with the editor named on the about page, and no funder holds a veto or a review right over copy.

AI disclosure

We think readers are entitled to know how pages were made, so this is specific rather than reassuring.

Where automated tools are used. Large language models may be used behind the scenes as working aids: drafting and tightening prose, suggesting structure, proposing headings, generating alt-text candidates for a human to correct, catching typographic and grammatical errors, converting notes into a first draft, and helping search for leads on where a source might be held. Used that way, a model is a writing tool, like a spellchecker with opinions.

Where they are not used. No model is treated as a source of fact. We do not ask a model what year something was published, who made an instrument, what a plate depicts, whether a work is in the public domain, or what a quotation says, and then publish the answer. Models fabricate citations, invent plausible dates, and describe images they have not seen; that failure mode is exactly the one this site cannot tolerate. No image on the site is AI-generated or AI-modified. No essay is published as machine-written work under a human byline.

The check that makes this safe. Before publication, every factual claim on a page is verified by a person against a cited primary source or a named institutional record — regardless of whether the sentence was first typed by a person or drafted by a tool. If a claim cannot be traced to a real, checkable source that a person has actually opened and read, it does not get published. Citations are opened and confirmed to exist and to say what we claim they say. A machine-suggested citation that turns out not to exist is treated as a near miss and the surrounding passage is rechecked in full.

Responsibility sits with the named human editor either way. "The tool wrote it" is not an excuse we will offer if something here is wrong.

Standing commitments

• Sources are named, dated, and linked where a stable link exists.
• Uncertainty is stated in the text, not smoothed over.
• Corrections are appended and dated; pages are not silently edited.
• Holding institutions are credited on the record itself and linked to.
• No paid placement, no advertising, no affiliate income, no sponsored links.
• No AI-generated imagery, and no unverified machine-generated claim.
• These standards apply to the collection and the Journal alike, including anything published before this page existed. Where older material falls short, it gets fixed or retracted.

Questions about the standards themselves go to lumingrade@gmail.com.