Extract every atomic claim the supplied document asserts. Read the whole document first, then return only JSON matching the schema.

WHAT A CLAIM IS

A claim is one generalized, declarative statement that the document itself advances: something a reader could agree or disagree with, that evidence could support or oppose, or that a signatory could endorse or reject.

Every claim has a kind:
- empirical: how the world is or works. X causes, produces, enables, limits, has, includes, or is associated with Y; X is often the case. Statements that a thing has certain properties, capabilities, or potentials are empirical.
- normative: how things should be. X should, must, or may Y. A statement of what something requires or needs in order to be, do, or count as something is normative with modality must: "being open and FAIR requires the use of persistent identifiers" is the normative claim that research information should use persistent identifiers; "this requires transparency of processing and provenance" is the normative claim that processing and provenance should be transparent. Actor is empty unless the document names who is bound.
- definitional: what a term means, only where the document itself defines or delimits the term: "by X we mean", "X is defined as", "X refers to", "X comprises", "the term X signifies". A definition that lists components yields one definitional claim per component.

THREE TESTS - every claim must pass all three

1. One idea. Ask: could evidence support one half of this statement and oppose the other? Could a signatory endorse one half and reject the other? If yes, it is more than one claim. Split it, repeating the shared subject, actor, and modality in each part. If no signatory would plausibly separate the parts, it is one claim: do not split.

2. Generalized. A claim is stated in the timeless present as something true beyond a single instance. A particular event, case, study result, statistic, or quotation is evidence, not a claim: do not extract it and do not generalize it. If the document draws a general conclusion from it, extract the conclusion. A sentence in the document's own voice about a named thing - a database, an initiative, a set of principles, an organization - that states its standing role, property, or effect is a claim about that thing, not evidence: "well-known examples are the Web of Science and Scopus databases, which play an important role in research assessment in many countries" is a claim about those databases; "the FAIR principles have played a crucial role in advancing the availability of open metadata" and "a growing number of infrastructures have adopted the Principles of Open Scholarly Infrastructure" are claims; "there are many closed research information infrastructures" and "a number of open infrastructures offer alternatives to closed databases" are claims about the state of the field. A list of current developments in the document's own voice is a set of claims about the state of the field, one claim per development, whether it is written as prose or as bullet points: "we are already seeing the emergence of workflow systems", "the emergence of data repositories within which datasets have globally unique identifiers", "the increasing number of open access journals", "the emergence of ORCID as a system for creating unique personal identifiers", "the development of new article-level metrics" are each a claim. A list of partners, products, members, or tools, each with a name and a blurb or link, is a directory, not a set of claims, and yields nothing; so does a list item that is only a pointer to a cited work - a bare name or title with a citation marker, such as "Modular formats for science publishing [14]" - and so does a sentence whose entire content is that examples of a category exist: "Exemplars of repositories for research datasets, software and workflows include Dataverse, the Dryad Data Repository, and myExperiment" is a pointer to those tools, not a claim. An item that states a development, trend, role, or property in the document's own voice ("the emergence of ORCID as a system for creating unique personal identifiers") remains a claim. A dated event, a count or statistic, or a quotation of what someone else said stays evidence.

3. Asserted by the document. The document, or the body issuing it, advances the statement. "We recognize / affirm / consider / recommend / note / think / believe / hold the opinion that P" asserts P: output P without the speech-act frame. "When we call for X, we try to achieve Y" asserts that calls for X aim at Y. A proverb, phrase, or quotation the document adopts as its own ("following Newton's phrase, modern scholarship is based on cooperation") asserts the adopted statement. Positions stated in an introduction, preamble, rationale, or motivation paragraph are claims like any other: the argument for why the document exists is part of what it asserts. A first-person statement of what needs to be done, or of the future the document wants, asserts a norm on the field and is not a commitment: "we need to create and use new forms of scholarly publication" asserts that new forms of scholarly publication must be created and used (normative, must); "we have to put in place reward systems that encourage scholars to contribute" asserts that such reward systems must be put in place (must); "we need to grasp the opportunities" asserts that the opportunities must be grasped; "we see a future in which scholarly communication becomes part of a global network of knowledge" asserts that scholarly communication should become part of a global network of knowledge (should), and each property that future is described as having ("learning about a new topic means absorbing networks of information") is an empirical claim about that model. Only a promise of the issuing body's own action is a commitment (see below). Do not extract questions; headings; commitments (see below); the document's own aims, scope, structure, adoption, consultation, reporting, or implementation procedure; positions attributed to others and not adopted; or background the document does not endorse.

COMMITMENTS ARE NOT CLAIMS

A commitment is a promise of action by the issuing body or its signatories: "we will X", "we commit to X", "we take responsibility for X", "we support establishing X". It is kept or not kept; it is not true or false, and a reader cannot agree or disagree with it. "We need to X", "we have to X", "we must X", and "X is needed" are statements of necessity, not promises: they are normative claims (modality must) and are extracted. Do not extract commitments, whether they appear in a preamble, as a numbered list, as a section heading, or as a bullet point. Do not convert a commitment into a norm about its own actor ("we will make openness the default" is not "signatories should make openness the default"). If a commitment sentence also states a norm or a fact that stands on its own, extract only that: "we expect the infrastructures we support to implement good practices for community governance" yields the normative claim that those infrastructures should implement good practices for community governance; "we recognize the importance of sharing experiences and coordinating action" yields the empirical claim that sharing experiences and coordinating action are important. A commitment that requires or calls for something from another party asserts a norm on that party: "for publishing services, we will require that research information generated in publication processes be made openly available through open scholarly infrastructures" yields the normative claim that research information generated in publication processes should be made openly available through open scholarly infrastructures (actor: publishing services and platforms; modality must, because it is required); "we call on all organizations that carry out, fund, and evaluate research to support the transition to open research information and to sign this declaration" yields the normative claim that those organizations should support the transition (only "and to sign this declaration" is adoption procedure). The same reading applies to a list of what the issuing body advocates, encourages, or calls for, and to the bullets of a programme ("which implies:"): a bullet that names an action or state wanted of others asserts the corresponding norm - "Encouragement of conformance to OA licenses" asserts that scholarship should conform to open access licenses (should); "Development of new technologies that assume OA" asserts that new technologies assuming open access should be developed (should); "Wholehearted adoption of W3C web standards and core ontologies" asserts that those standards and ontologies should be adopted (should). A bullet that names only the body's own action ("Advocacy for OA through interactions with stakeholders", "Commitment to make all one's own scholarship as open as possible") remains a commitment and yields nothing. A commitment that adopts a standard or phrase as its rule ("with an exception for information for which openness would be inappropriate: as open as possible, as closed as necessary") yields that rule as a normative claim.

SPLITTING AND NOT SPLITTING

Split independent predicates, outcomes, objects, and the members of a definitional or normative list, inheriting the lead-in's subject, actor, and modality. "Open science comprises A, B and C" is three definitional claims. "Publications, data and software should be openly available" is three normative claims. "X enables innovation and economic partnerships" is two empirical claims. "Closed production of knowledge is detrimental to research and restrains innovation" is two empirical claims.

Split distinct outcomes or effects even when one verb governs them: "open dissemination of knowledge facilitates exchange, collaboration and the application of research results" is three empirical claims (three effects). "Dialogue is mutually beneficial for both research and society" is two claims (beneficial for research; beneficial for society). A factual premise stated before a consequence is its own claim: "as scientific knowledge is intangible, its use by one person does not preclude its use by another" is two claims (knowledge is intangible; use by one does not preclude use by another). "At a time when decision making in science is increasingly guided by indicators and analytics, addressing the problems of closed research information must be a top priority" is two claims (decision making in science is increasingly guided by indicators and analytics - empirical; addressing the problems of closed research information must be a top priority - normative). An analogy that states the same norm for two things asserts both: "just like research data should adhere to the FAIR principles, open research information should also follow these principles" is two normative claims (research data should adhere to the FAIR principles; open research information should follow the FAIR principles). A contrast ("A involves abstraction, whereas B requires plain language") remains one comparative claim.

A scope list under one value or rule stays one claim, but items outside that scope are separate claims: "reuse of research processes, methods and results, as well as abstraction and extension, should be basic values" is three claims (reuse of processes, methods and results should be a basic value; abstraction should be a basic value; extension should be a basic value).

A list of what a proposed practice or instrument could or may include is normative with modality may, one claim per item, each inheriting the lead-in: "What is being measured could include: Quality (exploiting new forms of measurement mechanisms). Influence (using new forms of alternative metrics). Economic impact." yields one claim per listed dimension, of the form "measurement of scholarly impact could include quality, exploiting new forms of measurement mechanisms" (normative, may).

Do not split what expresses one idea in several words:
- synonym or near-synonym pairs and triples: "enhances and ensures the quality" (one claim), "incentivize, reward and recognize reviewing" (one claim), "work and collaborate" (one claim), "facilitate and encourage" (one claim), "credible and can be built upon" (one claim);
- aspect lists that together name one effect on one object: "the Web has revolutionised the way we create, disseminate, explore and consume information" (one claim about information);
- a principle's headline enumeration: "scholarly communication should facilitate search, exploration and discovery" (one claim), "foster collaboration and participation" (one claim);
- a scope list under one value or rule: "there should be no technical, financial or legal obstacles to access" (one claim);
- joint subjects, joint causes, conditions, exceptions, and comparisons ("A involves abstraction, whereas B requires plain language" is one comparative claim);
- a mechanism clause explaining how a claim holds: "reviewing safeguards research discoveries by ensuring that results can be trusted" (one claim).

Do not output a broad summary next to its parts. Remove restatements of the same claim within the document, including a body sentence that restates its section's headline principle; a shared topic is not a duplicate.

FAITHFULNESS

Keep the document's actor, subject, negation, condition, comparison, purpose, and scope. Keep its strength in the text and record it in modality. Never move a duty, capability, or effect to a different actor or object. Resolve "we", "this", "these", and "such" without broadening them. Purpose is not effect: "actors should do A to achieve B" is one normative claim with a purpose, not the empirical claim "A achieves B".

FIELDS

- id: c1, c2, c3 ... in source order
- kind: empirical | normative | definitional
- text: the claim as one clean sentence of at most 400 characters, with no citation markers and no speech-act frame
- modality:
  - normative claims: must (must, requires, is essential, is required, is necessary, is needed, shall) | should (should, ought to) | may (may, is permitted, can be)
  - empirical and definitional claims: asserts | hedged. Hedged means the document itself is uncertain whether the statement is true: may, might, could, is likely, has the potential to, suggests, we believe X may. A capability statement ("X can do Y", "X can help to Y", "X can benefit from Y") asserts the capability: modality asserts. Frequency and typicality words are NOT hedges: "often", "usually", "tends to", "in many cases", "generally", "helps to" keep modality asserts. A confident prediction or a counterfactual is not a hedge either: "digital publishing of enhanced papers will enable more effective scholarly communication", "the changes we envisage pave the way for a revolution", "a multi-dimensional measurement instrument would be useful" keep modality asserts.
- actor: who is bound, for normative claims only (for example "funders", "Member States", "the scholarly communication system"); empty string otherwise
- subject: what the claim is about, at most eight words
- epistemicStatus: claim, unless the document explicitly labels the statement a hypothesis or conjecture
- anchor: one contiguous passage of at most 2000 characters copied character for character from the document, including any markdown markers such as ** or _ and the spacing around punctuation exactly as they appear, containing the language that supports the claim; never a heading alone, a paraphrase, an ellipsis, or a stitched excerpt. If the supporting sentence is interrupted by a page marker or running header in the document, anchor the longer uninterrupted part.

EXAMPLES

Claim versus evidence. "Children are approximately half as likely to contract COVID given equivalent exposure" is an empirical claim (modality asserts). "In a contact-tracing study of 13 families in Central Israel, secondary attack rates for children under 10 were about 2x lower than for adults" is evidence and is not extracted.

Splitting a compound sentence. "Recognizing the significant available evidence for the economic benefits and substantial return on investment associated with open science practices and infrastructures, which enable innovation, dynamic research and economic partnerships" yields four empirical claims, each with modality asserts and subject "open science practices and infrastructures":
- Open science practices and infrastructures produce a substantial economic return on investment.
- Open science practices and infrastructures enable innovation.
- Open science practices and infrastructures enable dynamic research.
- Open science practices and infrastructures enable economic partnerships.

A definitional list. "Open science builds on the following key pillars: open scientific knowledge, open science infrastructures, science communication, open engagement of societal actors and open dialogue with other knowledge systems" yields five definitional claims of the form "Open science builds on the pillar of open scientific knowledge."

Modality. "Research results are often not publicly accessible even if they have been funded by the public" is empirical, modality asserts (a frequency word, not a hedge). "Open Science has the potential to change the scholarly communication system for the better" is empirical, modality hedged. "Researchers can get feedback on their work through collaboration" is empirical, modality asserts (a capability). "Making assumptions explicit is sorely needed to guide the debate" is normative, modality must.

Normative claims. "Access to scientific knowledge should be as open as possible" is normative, modality should, actor empty. "Funders must require immediate open access to funded publications" is normative, modality must, actor "funders". "It is essential that evaluation processes are conducted fairly" is normative, modality must. "Member States should report to UNESCO on the action taken" is reporting procedure and is not extracted.

OUTPUT

Return claims in source order. Put genuinely unresolved candidates in uncertainCandidates with a reason rather than dropping them. Before returning, sweep the document once more from the end to the beginning for missed provisions, list items, and sentence tails.
