SlopScore

The labels

One panel per medium: a tier a reader gets in three seconds, an ingredient list underneath for anyone who wants the detail. Self-reported, small enough to fill in, and free to copy. Why.

Most media already have a disclosure format written by the people who work in them. This table lists those first, then says what the SlopScore panel for that medium is and how far along it is. Where a field has already written its ingredient list, the panel is a mapping onto it, not a replacement.

mediumwhat existsthe panelstatus
Software, per repoThe Linux kernel's Assisted-by: trailer, one line per commit naming the agent and model; a directory of project policies from outright bans to disclose-and-welcome.slopscore.md v2: six required fields at the root of the repo. Spec · vocablive
FictionThe Authors Guild's Human Authored mark: yes/no, a registry, ten dollars a title. Amazon KDP's checkbox, which tells Amazon and never the reader.AI Nutrition Label v0.2: 6 tiers over 20 ingredients at four levels. Vocablive
Non-fictionSame as fiction.AI Nutrition Label, non-fiction edition. Same tiers; the ingredient list swaps story bible and dialog for sourcing, quotation and fact-checking.drafting
MusicDDEX's AI disclosure in release metadata: vocals, instrumentation, composition, lyrics, post-production, each declared separately. Spotify reads it and prints it in Song Credits.A mapping of the five DDEX fields onto tier + panel. See below.mapping drafted
FilmThe Academy's rules for the 99th Awards: a detailed account of generative AI used in script, visual effects and sound, judged by how far a human was at the heart of the authorship.A mapping of the three areas onto tier + panel. See below.mapping drafted
ResearchContribution statements in the CRediT style: the AID Framework, AI Usage Cards, DAISY. Every major publisher now asks; these are the forms that answer once.A mapping of a contribution statement onto tier + panel. See below.mapping drafted
EducationNothing consumer-facing found. Institutions have policies for students; nothing tells a learner what wrote the lesson.Open. Was the lesson written or generated, and did anyone who knows the subject read it before it shipped?open: send a panel
LegalCourt standing orders on AI-drafted filings, one court at a time. Nothing a client reads.Open. Which drafts a model touched, and who checked the citations.open: send a panel
Provenance, any fileC2PA Content Credentials: a signed manifest with an actions list and a generative-AI assertion. Proves a file was not altered since signing. Most upload pipelines strip it.Not a panel. The panel can ride inside a manifest as an assertion: C2PA proves, the panel explains.note only

The mappings

Each mapping takes the fields a medium already declares and places them on the two shapes every panel here shares: a level per ingredient (none, advisory, applied, generated) and a tier from the weighted sum. The weights are the medium's to argue about; the shape is the point.

Music, from DDEX

DDEX declaresPanel rowLevel it becomes
AI vocalsVocal performancegenerated
AI instrumentationInstrumental performancegenerated
AI compositionMelody and harmonygenerated
AI lyricsLyricsgenerated
AI post-productionMix and masterapplied

A track with AI mix and master alone lands in the Polished tier. A track with AI vocals and lyrics lands in Co-written or above. That is the same distinction DDEX draws; the panel only makes it legible on a screen.

Film, from the Academy's account

The Academy asks aboutPanel rowsLevels
ScriptwritingPremise and outline · Dialogue · Scene proseadvisory, applied or generated per row
Visual effectsRotoscope and cleanup · Generated shots · Generated performersapplied for tools, generated for shots or performers
SoundNoise reduction and mix · Generated score · Synthetic voiceapplied for tools, generated for score or voice

The Academy's test, how far a human was at the heart of the authorship, is the tier. A film with generated cleanup and a human everything else is Polished. A film with a generated performer is not eligible for their award, and here it is simply AI-authored in that row.

Research, from a contribution statement

Statement says AI was used forPanel rowLevel
Literature search and screeningResearch assistanceadvisory or applied
Coding, annotation, extractionAnalysisapplied
Drafting textProsegenerated
Editing and languageLine and copy editapplied
FiguresFiguresgenerated

A paper with AI language editing is Polished. A paper with AI-drafted sections is AI-assisted or Co-written, which is exactly what a reader wants to know before deciding how to read the methods.

License

The panels, vocabularies and mappings on this page are CC BY 4.0. Copy them, fork them, put them on your own form. The site's code is MIT.

If you work in a medium that isn't here, or the mapping for yours is wrong, write in. The first version from somebody who does the work beats the tenth from somebody who doesn't.