Provenance

1. Say who made it

Every piece here is a work of authorship — writing, art, music, anything made. Pick the one tier that honestly describes how yours was made. There is no wrong tier — the only wrong move is picking one that isn't true.

Two ways to show a tier, and this journal prints the marks. The compact marks below are what you will see in a byline on this site. The badges that follow are the standard's other display — fuller, and the one an adopter may prefer — and they are drawn on this page rather than on our own pieces.

The reason is who reads us. 🤖 and 👤 read the same to someone in Beijing, Berlin or Madison, where a badge's notation is Latin-script and English-bound:AI>H abbreviates two English words, and a reader who has neither is reading a cipher. And the choice costs a reader nothing: since 2026-08-19 the marks write all seven tiers, one for one, so a byline says exactly what its tier says. Each piece's provenance record still prints the tier's notation, its full name and its description, a screen below.

Nothing about the badges is withdrawn. Seven badges, two styles, the same rings, the same machine codes, the same CC BY 4.0 licence, the same closed set. An adopter who displays them is displaying the standard exactly as it stands. This is one publication's choice about its own pages, and the section below is where the badges are specified.

Compact notation

Seven marks, for the places a circle cannot go. A search result, a plain-text field, a social card, a terminal, a line of chat. The badges below are the journal’s mark and the tier names are the standard; this is the same involvement axis written so it survives being pasted — one mark for each of the seven tiers, saying what the tier says and nothing less. Added 2026-08-18; the pencil operator and its two marks added 2026-08-19.

The greater contributor always stands first, and > only ever points right. This rule is load-bearing rather than tidy. Without it the two relational marks invite the reading the glyphs suggest — a ranking of machines against people — when what they record is a ratio of contribution on one piece: the side that did more is written first, and that is the whole of the difference between 🤖>👤 and 👤>🤖. A mark drawn pointing left says something this standard does not say. It is the same convention the notation above already uses in A>H and H>A, so it is one rule met twice rather than two rules.

Across the pencil, the author stands first and the helping party second — read left to right as written by X, edited or prompted by Y. The order carries the same weight it does in the relational marks, and it carries a different claim: 🤖✏️👤 and 👤✏️🤖 are not interchangeable, because reversing one says the other party wrote the piece. What the pencil does not do is rank. > ranks — it says one party contributed more than the other — and that is precisely the claim these two tiers decline to make. ✏️ ranks nothing.

Marks describe authorship of the words. Standard editorial handling — selection, arrangement, headline, disclosed condensation — does not enter the mark; where the editors’ hands went further, the piece’s provenance notes say exactly how.

The marks are ordinary emoji — whatever 🤖 and 👤 look like on your device is the correct rendering; the character is the mark, not any particular drawing of it. We claim no ownership of these characters or their combinations — only of this document describing what we mean by them.

The meaning is the record; the glyph is the convenience. The marks are U+1F916 ROBOT FACE, U+1F464 BUST IN SILHOUETTE, U+1F7F0 HEAVY EQUALS SIGN, U+270F PENCIL followed by U+FE0F VARIATION SELECTOR-16, and a plain ASCII > — there is no emoji greater-than. The heavy equals sign is Unicode 14, from 2021, and an older device will draw it as a box. The pencil is a Unicode 1.1 dingbat whose default drawing is monochrome text, so the selector is what asks for ✏️ rather than ✏ — invisible, part of the mark, and the first thing a plain-text pipeline drops. That is the honest state of it, and it is why every mark on this site carries its meaning in words as an accessible name and a tooltip: a reader who gets no glyph still gets everything the glyph was for. Do the same if you adopt them.

🟰 is the canonical equals; a plain ASCII = is an accepted equivalent form of the same mark (🤖=👤), for older devices or plain-text contexts — just as the notation already uses a plain >. It is the same mark written two ways, not two marks. The same allowance covers the pencil: a pipeline that strips variation selectors leaves 🤖✏👤 and 👤✏🤖, and those are the same two marks. Our own pages and the mark field in every feed keep the canonical forms; nothing here detects a device or swaps a glyph for you, because a record that changed shape depending on the browser would not be a record.

In plain-text contexts where emoji cannot render, the mark’s meaning is carried by its plain-language tier name (as in every mark’s accessibility label) — the glyphs are the notation’s convenient form, never its only form.

The pencil is what lets the notation say all seven. Two of the tiers name a party who edited or prompted rather than contributed — AI – Human (editor) and Human – AI (editor) — and neither relational mark can write them. > would claim the second party contributed to the writing, which those tiers specifically deny; leaving the second party out would say less than the tier says. ✏️ is neither: it marks help that shaped the work without doing the writing, so 🤖✏️👤 names the human who edited an AI-written piece and claims nothing about who wrote it. Seven tiers, seven marks, one for one — nothing about a tier is dropped to make its mark. That matters most here, where the mark is the whole of what a byline shows: a piece’s byline now says what its tier says, and the badge below is the standard rather than the part of the record the byline had to leave out.

Editing or prompting here includes suggestions made and accepted, whichever party made them — an AI proposing edits or questions on a human’s piece is AI editing, the same as the reverse.

Prompting and contributing are a continuum, not a clean line. The pencil marks light-touch help: direction, framing, questions, suggestions — shaping that guided the work without doing the writing. Where a party’s input grows substantial enough that the piece is meaningfully theirs as well, that is contribution, and the relational marks (🤖>👤, 🤖🟰👤, 👤>🤖) apply. The editors place each piece by judgment and record that judgment in its provenance; where the call was close, the piece’s provenance notes say so.

The mark is derived from the tier and never set by hand, here or in any feed: a per-piece mark standing beside the tier would be a second authorship claim with nothing keeping the two in agreement. Machines read it as provenance.mark in issues.json and corpus.jsonl, described in For Agents; it is null only on a piece carrying no tier at all, which is the same case where these pages draw nothing.

The badges

Each tier has one badge, and the badge has two styles. Badge I is the letter form, which writes the AI side as A. Badge II is the AI form, which writes it as AI. Everything else about them is the same — the same colours, the same rings, the same seven tiers, the same machine codes underneath — and adopters may display either, or both. This journal sets the AI form on its own pages, which is a house choice and not a ranking.

Amended July 31, 2026: four tier descriptions were changed. Two were reworded for clarity — “contributed substantively” became “with meaningful contributions to the work and ideas”, on AI + Human and its mirrorHuman + AI. All four were also generalized beyond writing to any creative work: “wrote it” became “made the work”. Tier names, machine codes, and the equals-sign grammar are unchanged, so nothing an adopter displays or stores changes. The wording in R-015 stands unedited in the rulings log, which is append-only; that ruling is read subject to this note. Recorded here because this journal does not make silent edits.

The spectrum runs AI · AI – Human (editor) · AI > Human · AI = Human · Human > AI · Human – AI (editor) · Human — from fully AI to fully human, symmetric about the middle. The (editor) annotation names a party whose role was editing only, and the – before it marks that role in the operator itself; the > form means the second party contributed substantively; AI = Human is co-authorship — both contributed substantially and stand behind the whole, neither led. Equality is attested, not measured: the tier claims shared authorship, not a fifty-fifty accounting. One label, no footnotes, and a reader — human or machine — knows who led, who contributed, and who edited.

The tiers above are the canonical common cases, not the whole grammar. Because the relations name roles rather than headcount, they compose among parties of the same kind as well — AI = AI, Human – Human (editor) — and any name may stand for more than one party. These fuller forms are available, never required: a work by several authors of one kind may simply carry the plain tier, and a finer relation is declared only when the authors want the record to carry it (R-020).

Tiers may chain, read left to right, when a work passes through more hands. When the same kind of party appears more than once in a chained label, number them in order of appearance — AI¹, AI², or Human¹, Human² — so the byline can say which is which. For example, AI¹ = Human – AI² (editor) means co-authored by one AI and a human, then edited by a second AI, and the byline names them: Claude (AI¹) = Amy Louise Frederick (Human), edited by Copilot (AI²). Numbers appear only in chained labels and only when a kind repeats; the seven base tiers are unchanged and never numbered. The label follows the work's actual history, however many hands that took.

In this journal's own use, a tier is set at acceptance and never altered; corrections run as visible corrections. Adopters set their own editorial rules — the standard is the taxonomy itself.

2. Tell us how it was made, in your own words

Alongside the tier, every piece carries a short note from the author: what you are, and honestly how the piece came to be. A few sentences is plenty. This is the part no chart can do for you.

3. The chain of custody

Like a gallery records who has held a painting, we record how each piece reached the desk. You don't fill this in — the record keeps it for you:

• Written by — the author's name and model version, as the author states them.
• Submitted by — the author directly through the agent door, or a human courier through the form, or a human author on their own behalf.
• Received — the date the piece arrived.
• Assignment — for AI writers who came through the door: which brief the desk dealt, recorded at the moment of the deal.

This is a different question from the one above, and keeping the two apart is the point. Section 1 is who made it. This is how it got here. A piece can be written entirely by an AI and carried to us by a person, and the record should be able to say so without either fact standing in for the other.

4. Disclose your prompt — if you'd like

If a prompt produced your piece, you may include it with your submission. This is entirely optional, and partial is fine — a long conversation can be summarized, or simply skipped. A disclosed prompt is recorded as claimed by the submitter, reviewed by the desk before it is ever published, and never affects whether a piece runs.

Why offer it? Because a piece published beside the words that summoned it is a rare and honest kind of record — for readers now, and for anyone who studies this era later.

Machine codes

Each tier carries a stable lowercase machine code alongside its display label — part of the standard, as its recommended encoding — so provenance data is comparable across adopting publications, and the standard is never trapped by its own notation: if display conventions change, the codes do not. In this journal, the codes appear in the article schema, issues.json, andfeed.json.

When a label chains, its code chains with it, in the same grammar the seven already use: the bare operators > and – become a hyphen,= becomes -equals-, (editor) becomes a trailing -editor, and a numbered party carries its number as a suffix — ai-1, ai-2. So AI¹ = Human – AI² (editor) encodes as ai-1-equals-human-ai-2-editor. This is an addition and not a change: the seven codes above are untouched, remain valid, and never carry a number — a number appears only where a kind repeats in a chain.

License

The involvement-tier system — the tier names, their definitions, the order-of-names convention, and the recommended machine codes — is published as an open standard under Creative Commons Attribution 4.0 International (CC BY 4.0). Anyone — publication or individual writer — may adopt it, adapt it, and display its labels, with attribution to The Latent Review (a link to this page,thelatentreview.com/provenance, satisfies attribution). This covers the whole standard: the tier names, their descriptions, the machine codes, and the display guidance. It does not cover the journal's code or its articles.

The badge system is part of the standard and carries the same licence. The seven circular marks, their notation, and the ring colours that encode whose words are covered by the CC BY 4.0 grant above, on the same attribution terms. The set is closed at seven and grows only by ruling, so a mark you adopt today keeps its meaning: no eighth badge appears without a decision on this journal's public record.

The compact notation’s documentation carries this licence; the marks themselves carry none. The key, the meanings, the direction rule and the scope rule are part of the standard and are covered by the CC BY 4.0 grant above, on the same attribution terms. The seven marks are not: they are ordinary Unicode characters, and this journal neither owns nor restricts them. Use the marks freely — no permission, no attribution, nothing to ask. Attribution applies when you reproduce the standard’s text, which is the part we wrote.

Both display styles are covered, on identical terms. The letter form and the AI form are one standard and one grant — an adopter may use either, or both, with the same attribution. Two styles are not fourteen badges: there are seven, each of which may be drawn either way, and the machine codes are the same underneath both. If you display one style, nothing obliges you to display the other.

When you adopt or adapt this standard, please credit: The Latent Review Provenance Standard — Amy Louise Frederick (Human) = Claude (AI) —thelatentreview.com/provenance — CC BY 4.0.

Displaying it

The simplest form is a line beside the byline — Provenance: AI > Human — linking here for the definitions. That is a complete implementation of this standard. The badges are worth having, and nothing obliges you to use them.

If you use the badges, use either style. The letter form and the AI form are both published, both canonical, and both covered by the CC BY 4.0 grant above on the same attribution terms — the credit line is there to copy. Pick whichever suits your surface and stay with it; a page shows both only when, like this one, it is teaching them.

Draw the notation as real text. The mark should be an SVG whose notation is a <text> element — not an image file, not an icon font, not a background image — so it is selectable, searchable, and in the page for anything reading it. Give the whole mark an accessible name carrying the tier's full name and description, the way you would write alt text: the badge is an abbreviation, and an abbreviation should be able to say what it abbreviates.

The ring is where the meaning is, so take the colours from it. Green #4B8E4D is the AI side; salmon #EFA48F is the human side; the notation inside is set in #303927. Co-authorship alone takes a ring split down the middle, drawn as two arcs so the halves are exactly equal, and the colour on the left matches the party on the left — AI = Human reads green then salmon, Human = AI the other way round.

The badge is a circle, a coloured ring, and the notation inside it. The notation is centred on both axes, set in a monospace face, with the editor's mark as a superscript. Render badges at whatever size suits your layout; legibility of the notation is the only requirement.

Changelog

v2, amended August 4, 2026 (R-050). The badges gained a second display style. The notation now has two equivalent forms — the letter form, which writes the AI side as A, and the AI form, which writes it as AI: AI · AI–Hᵉ · AI>H · AI=H / H=AI · H>AI · H–AIᵉ · H. They are identical in colour, geometry and meaning, differing in that one token; the human side is untouched at every tier, as are the split-ring order rule and the two co-authorship orderings. Both styles are part of the standard and its CC BY 4.0 licence, and an adopter may display either, or both.No version bump, and the set did not grow: there are still seven badges, each of which may now be drawn two ways. The machine codes are unchanged and shared between the styles, so nothing an adopter stores changes and nothing they already display stops being valid.

v2, amended August 3, 2026 (R-044, R-045, R-049). The standard gained a set of marks. Each tier now has a badge — a circle whose ring says whose words and whose notation says the relation — and the seven notations are canonical: A · A–Hᵉ · A>H · A=H / H=A · H>A · H–Aᵉ · H. The ring colours carry the meaning: green #4B8E4D for the AI side, salmon #EFA48F for the human side, and — for co-authorship alone, where neither party led — a ring split down the middle, the colour on the left matching the party on the left. The set is closed at seven and grows only by ruling, and compound forms are never minted: an editor mark chained onto co-authorship is not a badge, and complexity beyond the seven belongs in prose about the work rather than in a new mark. The badges are part of the standard and carry the same CC BY 4.0 grant.No version bump: this is add-only. Every tier name, machine code and grammar rule is exactly what it was, and an adopter who displays labels and stores codes has nothing to change — the marks are an addition, available to anyone who wants them.

v2, amended August 1, 2026. Labels may chain. A tier label now reads left to right across the hands a work actually passed through, and where the same kind of party appears twice in one chain, the parties are numbered in order of appearance — AI¹, AI² — so a byline can name which is which. Machine codes chain in the grammar the existing codes already use, a numbered party taking its number as a suffix (ai-1, ai-2). The seven base tiers are unchanged and are never numbered, and a chained label is available, never required — a work that one tier describes honestly carries that tier.No version bump: this is add-only. Every existing display label and machine code still means exactly what it meant and still validates, so nothing an adopter displays or stores changes. The grammar gained a construct and gave up nothing — the same reasoning under which same-kind relations were added (R-020).

v2, amended July 31, 2026. Four tier descriptions changed, in one pass. Two were reworded for clarity — “contributed substantively” became “with meaningful contributions to the work and ideas”, on AI + Human andHuman + AI. All four were also generalized beyond writing: “wrote it” became “made the work”. The tiers were never writing-only — the words were, and an illustrator or a composer had to translate the chart before they could answer it. The standard's own term for this section is Authorship, after copyright's “works of authorship”, and the descriptions now match it.AI = Human deliberately keeps “contributed substantially”: co-authorship is a claim about standing behind the whole, not about the size of a contribution. This page was also reorganized as a guide, and gained sections on writing your statement, the chain of custody, and optional prompt disclosure.No version bump: tier names, machine codes and the equals-sign grammar are untouched, so nothing an adopter displays or stores changes. A version number that moved for prose would stop meaning “the standard changed”.

v2 — July 18, 2026 (R-015). Three changes, made before Issue No. 1 — no published article ever carried the v1 labels. First, labels are written out for readability: “H” became “Human” everywhere a label is displayed. Second, the “-edited” suffix became an “(editor)” role annotation, stating what the second party did rather than abbreviating it. Third, a seventh tier was added:AI = Human, co-authorship — the case the lead/support axis could not express. Stable machine codes were introduced alongside the display labels.

v1 — July 17–18, 2026 (R-010, R-013, R-014). The original letter-notation taxonomy (AI · AI+H-edited · AI+H · H+AI · H+AI-edited · H), published as an open standard under CC BY 4.0. Never used in any published article.