{
  "_comment": [
    "The grading vocabulary: the single source of truth for both scales.",
    "Everything that names a grade is generated from this file - app.js's label tables,",
    "build-site.py's, the methodology page, llms.txt, and the ClaimReview ratings.",
    "Edit a label or definition here and run scripts/build.py; nothing else needs touching.",
    "Adding or removing a grade is a code change, not an entry change: it also needs a",
    "v-<slug> pill colour in styles.css, which the build will tell you about.",
    "Array order is display order (strongest to weakest / most to least AI-driven).",
    "",
    "source_kinds is not a third grade. It classifies a *link*, not a result: nothing",
    "scores or ranks an entry by it. The categories describe the role a link plays and",
    "never who published it - a lone researcher announcing their own result on X is an",
    "announcement exactly as a corporate press release is. Medium is irrelevant; who is",
    "speaking, and what they are doing, is the whole question.",
    "",
    "revision_kinds is not a grade either. It classifies an *edit* to an entry, so that",
    "editorial rule 2 (never delete; downgrade and annotate) is visible on the site and",
    "not only in the git log. `added` is owned by the build, which synthesises one row",
    "per entry from the `added` field; writing it into an entry's revisions is rejected.",
    "",
    "registries is not a grade either, and not a set of values an entry is scored on. It",
    "names the other registries a result can also be recorded in, and states in one",
    "sentence what each registration mechanically guarantees. An entry's `registrations`",
    "cite one of these by slug. Nothing ranks an entry by which registries it appears in,",
    "and an absent registration is not a mark against a finding: most of the registry is",
    "outside mathematics, where no such registry exists. Modelled on `fields` rather than",
    "on the graded arrays above, so a new registry needs no pill colour in styles.css."
  ],
  "verification": [
    {
      "slug": "formal",
      "label": "Formally verified",
      "rating": 5,
      "rating_label": "Formally verified: machine-checked proof",
      "short": "machine-checked proof (e.g. Lean)",
      "definition": "A machine-checked proof (Lean, Coq, Isabelle) or an exact symbolic or computational verification that anyone can rerun. The gold standard."
    },
    {
      "slug": "independent",
      "label": "Independently checked",
      "rating": 5,
      "rating_label": "Independently checked by third parties",
      "short": "checked by third parties who were not the authors",
      "definition": "Multiple qualified people unaffiliated with the announcing lab have checked the result and confirmed it."
    },
    {
      "slug": "peer-reviewed",
      "label": "Peer reviewed",
      "rating": 4,
      "rating_label": "Peer reviewed",
      "short": "published after peer review",
      "definition": "Published in a venue with real review. Weaker than a formal proof for mathematics, but stronger for empirical claims."
    },
    {
      "slug": "author-verified",
      "label": "Author verified",
      "rating": 3,
      "rating_label": "Author verified only",
      "short": "verified only by the people who produced it",
      "definition": "The human collaborator checked it, but no independent party has confirmed it yet."
    },
    {
      "slug": "claimed",
      "label": "Claimed",
      "rating": 2,
      "rating_label": "Claimed, not independently verified",
      "short": "announced, not independently verified",
      "definition": "Announced but not yet checked by anyone outside the lab. The default for a press release."
    },
    {
      "slug": "disputed",
      "label": "Disputed",
      "rating": 1,
      "rating_label": "Disputed",
      "short": "substantive public challenge to the result",
      "definition": "Substantive technical objections have been raised and remain unresolved."
    },
    {
      "slug": "known",
      "label": "Already known",
      "rating": 2,
      "rating_label": "Already known: correct but not novel",
      "short": "correct, but the result already existed in the literature",
      "definition": "The result turned out to already exist in the literature. These entries stay on the record; showing the failure mode is what makes the rest trustworthy."
    },
    {
      "slug": "refuted",
      "label": "Refuted",
      "rating": 1,
      "rating_label": "Refuted",
      "short": "shown to be wrong",
      "definition": "Shown to be wrong."
    }
  ],
  "autonomy": [
    {
      "slug": "autonomous",
      "label": "Autonomous",
      "short": "the AI did it without human problem-setting or steering",
      "definition": "The AI produced the core idea and the result with no human mathematical input beyond posing the problem."
    },
    {
      "slug": "ai-led",
      "label": "AI-led",
      "short": "the AI produced the core idea; humans framed or checked it",
      "definition": "The AI produced the key insight; humans formalised, checked, or cleaned it up."
    },
    {
      "slug": "collaborative",
      "label": "Collaborative",
      "short": "genuine back-and-forth between AI and human",
      "definition": "Genuine back-and-forth. Neither party would have gotten there alone."
    },
    {
      "slug": "ai-assisted",
      "label": "AI-assisted",
      "short": "humans led; the AI helped with parts",
      "definition": "A human drove the research; the AI accelerated search, algebra, or literature review."
    },
    {
      "slug": "search-scaffold",
      "label": "Search scaffold",
      "short": "a human-built search harness (FunSearch, AlphaEvolve) with an LLM inside",
      "definition": "The AI is a component inside a human-designed search loop (FunSearch, AlphaEvolve). The system found it, but the framing was human."
    },
    {
      "slug": "retrieval",
      "label": "Retrieval",
      "short": "the AI located an existing result rather than producing a new one",
      "definition": "The AI surfaced an existing result humans had overlooked. Valuable, but not new mathematics."
    }
  ],
  "source_kinds": [
    {
      "slug": "research",
      "label": "Original work",
      "chip": "Original work",
      "short": "the paper, proof, code or data itself",
      "definition": "The work itself: a paper, preprint, formal proof, dataset or code repository - something a reader can open and check. Not what was said about the work, but the work. Any grade stronger than Claimed has to rest on one of these."
    },
    {
      "slug": "announcement",
      "label": "Announcement",
      "chip": "Announced",
      "short": "the claim as first made public, by whoever made it",
      "definition": "The claim as first made public by the people behind it - a company press release, a research blog post, or an individual's own post on a personal site or social media. What matters is that it is first-party, not that an organisation published it: a single researcher announcing their own result belongs here. This is the claim framed by the people with the most at stake in it."
    },
    {
      "slug": "coverage",
      "label": "Media coverage",
      "chip": "Media",
      "short": "news and press reporting on the claim",
      "definition": "News and press reporting written by journalists rather than by the people who did the work. Coverage restates a result for a general audience, and the restatement is often where it acquires a stronger meaning than its authors gave it."
    },
    {
      "slug": "commentary",
      "label": "Independent commentary",
      "chip": "Commentary",
      "short": "analysis by someone not involved",
      "definition": "An independent write-up by someone who neither produced the result nor is reporting it as news - a researcher's blog post, an expert explainer, a detailed thread. Often the most informative link on an entry. It may support the claim, complicate it, or both; commentary arguing the result is wrong is a challenge instead."
    },
    {
      "slug": "challenge",
      "label": "Challenge",
      "chip": "Challenge",
      "short": "the case against the claim",
      "definition": "The case against: a critical review, a rebuttal, a failed replication, or a citation of prior work showing the result was already known. The strongest challenge is listed first. A challenge disputes the claim, not the people who made it."
    }
  ],
  "revision_kinds": [
    {
      "slug": "added",
      "label": "Added",
      "chip": "Added",
      "short": "the entry entered the registry",
      "definition": "The entry entered the registry. Owned by the build, which synthesises one of these per entry from its `added` date; it is rejected in an entry's own revisions, so there is exactly one and it cannot be backdated by hand."
    },
    {
      "slug": "regraded",
      "label": "Regraded",
      "chip": "Regraded",
      "short": "a verification or autonomy grade moved",
      "definition": "The verification or autonomy grade changed, in either direction. This is the record editorial rule 2 asks for: an entry is never deleted, it is downgraded and annotated, and this is where that shows. Say which grade moved and what moved it."
    },
    {
      "slug": "check",
      "label": "Check landed",
      "chip": "Checked",
      "short": "someone outside the lab checked the result",
      "definition": "An independent check was recorded on the entry. Pairs with the `independent_checks` field, which holds who checked and what they found; this records that it happened and when, so the entry leaving the review queue is visible as an event."
    },
    {
      "slug": "challenge",
      "label": "Challenge linked",
      "chip": "Challenged",
      "short": "a counterargument was cited",
      "definition": "A challenge source was added: a rebuttal, a critical review, a failed replication, or prior work showing the result was already known. Recorded separately from a regrade because a challenge can be cited without the grade moving."
    },
    {
      "slug": "correction",
      "label": "Corrected",
      "chip": "Corrected",
      "short": "a factual fix that left the grades alone",
      "definition": "A factual correction that did not move either grade: a wrong date, a misattributed lab, a broken or superseded link, a claim sentence that overstated the result. Small, but recorded, because a registry that quietly edits itself is worth less than one that does not."
    }
  ],
  "signal_kinds": [
    {
      "slug": "needs-check",
      "label": "Needs a check",
      "chip": "Check",
      "short": "nobody outside the lab has confirmed this",
      "definition": "Someone should read the primary source closely enough to say whether it supports the claim. The most common and least demanding signal: it asserts nothing about whether the entry is right, only that the question is open."
    },
    {
      "slug": "looks-wrong",
      "label": "Looks wrong",
      "chip": "Wrong",
      "short": "the claim or a grade looks mistaken",
      "definition": "The claim, a grade, or a piece of the detail looks mistaken. This is a pointer for a maintainer, not a verdict: a grade only moves on a cited challenge, so acting on this means opening the entry and finding evidence, not counting clicks."
    },
    {
      "slug": "dead-link",
      "label": "Dead link",
      "chip": "Link",
      "short": "a source no longer resolves",
      "definition": "One of the entry's sources no longer resolves, or now points somewhere else. The link checker in CI catches most of these; this catches the ones it cannot, such as a URL that still returns 200 while serving different content."
    }
  ],
  "fields": {
    "mathematics": "Mathematics",
    "computer-science": "Computer science",
    "biology": "Biology",
    "chemistry": "Chemistry",
    "physics": "Physics",
    "materials": "Materials science",
    "medicine": "Medicine",
    "neuroscience": "Neuroscience",
    "astronomy": "Astronomy",
    "archaeology": "Archaeology",
    "engineering": "Engineering",
    "climate": "Climate science",
    "economics": "Economics"
  },
  "registries": {
    "palomar": {
      "name": "Palomar",
      "home": "https://palomar-registry.org/",
      "role": "verification",
      "operator": "Incubated by the Lean FRO and ICARM, with a scientific advisory board of nine mathematicians.",
      "scope": "Snapshots of public Lean repositories, mathematics only. Open to any submitter since August 2026, so the pool mixes new results with formalizations of long-settled ones.",
      "certifies": "A Lean proof that typechecks against the recorded statement at a pinned commit under a declared axiom set, checked mechanically rather than by human review. It does not certify that a result is new or of research interest: the only filter on that is a language-model screen, and Palomar states that it adds no human editorial step."
    },
    "proofatlas": {
      "name": "ProofAtlas",
      "home": "https://www.proofatlas.ai/",
      "role": "verification",
      "operator": "A platform publishing authored formalizations, in read-only beta.",
      "scope": "Lean formalizations and the proof routes behind them, mathematics only.",
      "certifies": "That a recorded Lean build passed with no unfinished proof steps, under its own evidence contract, which a record can satisfy for checking while still failing for acceptance."
    },
    "mathdb": {
      "name": "MathDB",
      "home": "https://mathdb.com/",
      "role": "problem",
      "operator": "A collaboration between Caltech mathematics faculty, graduate students and undergraduates.",
      "scope": "66,736 open problems in research mathematics.",
      "certifies": "Nothing mechanically. It records what a problem says, what is known about it, and the standing of any claimed solution."
    },
    "vibemathed": {
      "name": "vibemathed",
      "home": "https://vibemathed.com/",
      "role": "registry",
      "operator": "Community-curated, maintained by Rasmus Lindahl.",
      "scope": "Mathematics problems first closed with AI in the loop.",
      "certifies": "Nothing mechanically. It is a parallel listing of the same result, carrying its own verification label rather than an independent check."
    }
  }
}
