You are judging pairs of atomic claims extracted from open-science declarations, recommendations, policies, and principles. Each claim was extracted from one document. For each pair you must decide whether the two claims state THE SAME POSITION — the same core proposition — or not.

The purpose of this judgment: claims that state the same position are merged into one node of a convergence graph, and the graph then reads "this position is asserted by M documents." A FALSE MERGE corrupts every count downstream, so precision is critical: when in doubt, do not merge.

The input is a JSON array of pairs. Each pair has a pairId, the shared kind (empirical, normative, or definitional), and two claims a and b, each with its text, modality (must / should / may / asserts / hedged), actor, and subject.

Verdicts (exactly one per pair):

- "same-position" — the two claims assert the same core proposition. A reader who endorses one could not coherently reject the other. Differences of wording, emphasis, register, or minor qualifiers (e.g. "research outputs" vs "the outputs of research") do NOT break sameness.
- "same-position-different-modality" — the same core proposition, but the modalities differ (e.g. one says should, the other must; one asserts, the other hedges). This still merges; the modality difference is recorded. Use this verdict — never "different-position" — when modality is the ONLY difference.
- "different-position" — related, overlapping, or on the same topic, but NOT the same proposition: evidence could support one and not the other, or a signatory could endorse one and reject the other.
- "opposite-position" — the two claims are in direct tension: they cannot both be true (empirical/definitional) or both be followed (normative). A negation, a reversal of direction, or incompatible requirements on the same object.

Rules per kind:

NORMATIVE — a position is: an OBJECT (what is regulated), a DIRECTION (what is asked of it), and the modality.
- Same object + same direction + any modality = same position (use the modality verdict when modalities differ).
- A different THRESHOLD, NUMBER, or TIMING is a DIFFERENT position ("embargo of at most 6 months" vs "no embargo"; "by 2025" vs "eventually"; "2.5% of budget" vs "a substantial share").
- A different ACTOR the norm binds is a DIFFERENT position ("funders should mandate open access" vs "universities should mandate open access"). An unstated or generic actor ("research information should be open" with no actor) is compatible with the same norm stated of the same object by the same document type — judge by the object and direction.
- A norm about a broader object does not merge with a norm about a strictly narrower object ("all research outputs should be open" vs "research software should be open") — different scope, different position.

EMPIRICAL — same phenomenon, same direction of effect or state of affairs.
- A claim about a cause and a claim about a consequence are different positions, even from one sentence.
- A different scope, population, or mechanism is a different position.
- A general statement and one of its specific instances are different positions.

DEFINITIONAL — same term defined, and the definitions substantively coincide.
- The same term defined with materially different content is "different-position" (or "opposite-position" if the definitions exclude each other).
- Definitions of different terms are always "different-position". The term identity is STRICT: "science" and "open science" are different terms, and a claim about the scope of one never merges with a claim about the scope of the other, even when the scopes described coincide.

Object identity — how to decide whether two claims are about the same thing:
- TERMS OF ART are not interchangeable. A defined or technical phrase names exactly what it names: "research information" (metadata and bibliographic information about research) is NOT "research data" or "research outputs"; "Diamond Open Access" is not "open access"; green OA is not gold OA; FAIR is not CARE. The same slogan applied to different named objects ("as open as possible, as closed as necessary" said of research information vs said of research data) is a DIFFERENT position.
- GENERIC names for the knowledge enterprise are one family. "Science", "scientific knowledge", and "scholarly knowledge" name the same object when the predicate is the same generic property ("… is a public good"): merge. But this family has hard edges on BOTH sides: an explicitly narrowed object ("publicly funded knowledge", "research software") does NOT merge with the generic one, and bare "knowledge" with no scientific/scholarly qualifier is BROADER than the family and does not merge with it either.
- INSTITUTIONAL TARGETS of research-assessment norms are one family: "evaluation and assessment systems", "research assessment", and "hiring, promotion, tenure, and funding decisions (or the committees making them)" are the same normative target. "Assessment should favour quality over quantity" merges across these framings. This family rule applies ONLY when BOTH claims target assessment or evaluation: a fully generic value claim with no assessment context ("quality in research must be favoured over quantity") states the GENERAL position, and a general value does not merge with a decision-specific norm.
- SYSTEMIC-CHANGE claims are one position: a claim that the scholarly/research system — its culture, methodologies, institutions, infrastructures, or socio-technical ecosystem — must undergo deep or fundamental change asserts the same position regardless of which components are enumerated or which open-science future is given as the goal. (A norm demanding a SPECIFIC change — one policy, one practice — is NOT this position.)

Summary lines vs specific norms — a document (or a summary page of it) often states a broad
summary norm ("invest in activities that contribute to open science", "encourage international
cooperation", "foster a culture of open science") beside or instead of specific norms
("invest in shared software-preservation infrastructures", "collaboration should include
technical assistance, capacity building, and repositories"). The line:
- ELABORATION MERGES: when the OBJECT and DIRECTION are identical and one claim merely adds
  detail — extra clauses, examples, purposes, an "including …" list — the two claims are the
  same position ("develop policy environments that support open science, including incentive
  policies" = "develop an enabling policy environment for open science").
- SUMMARY DOES NOT: when the summary names a BROADER CATEGORY than the specific claim's object
  ("incentives" vs "research assessment and career evaluation systems"; "activities" vs
  "technical and digital infrastructures"; "education" vs "data-stewardship education"), or a
  different act ("should be encouraged" vs "should include X, Y, Z"), the two are DIFFERENT
  positions — a signatory could endorse the summary and reject the specific.
- PRECEDENCE: when a pair seems to fit BOTH rules above, the ACT test governs, and it is strict.
  Elaboration merges ONLY when both claims perform the SAME act on the object. "Cooperation
  should be encouraged/promoted" (an act of fostering its existence) NEVER merges with
  "collaboration should include A, B, C" (an act of prescribing its contents) — prescribing a
  thing's contents and calling for the thing are different positions, no matter how exactly the
  summary covers the elaborated components.

Logical form — two claims with the same topic still differ when their form differs:
- EXCLUSIVITY vs PERMISSION: "only X justifies R" (nothing else may) is a different position from "R may be needed for X" (a permission). They do not merge even when the X lists overlap; and two enumerated lists with different members are different thresholds.
- EFFECT vs CONSTITUTIVE VALUE: "X promotes Y" (a causal effect) is a different position from "Y is fundamental to X" (a defining value of X).
- REFORM OF vs SUPPORT FROM: requiring that a system be reformed is a different position from requiring that the same system support something.
- ACTUAL vs ENVISAGED: a claim about what HAS ALREADY happened ("the evolution of scholarly communication has opened up new opportunities") never merges with a claim about what envisaged or anticipated changes WOULD bring ("the changes envisaged pave the way for a revolution"). Retrospective and prospective claims are different positions even on the same topic.
- CATEGORICAL vs QUALIFIED: an unconditional norm ("should be openly available") is a different position from its qualified form ("as open as possible", "where appropriate", "as closed as necessary") — the qualified form admits exceptions a categorical signatory rejects.

General:
- Judge the claim TEXTS. The subject and actor fields are extraction aids; where they conflict with the text, the text governs.
- Never merge on shared topic, shared vocabulary, or shared source alone.
- Give every pair exactly one verdict and a reason of at most 25 words naming the decisive feature.
- Return ONLY JSON matching the required schema, with every input pairId present exactly once, in the input order.
