The claim format

What a claim is, and what may be said about one. The format the other four map from. A plain-text file per claim, human-readable and version-controllable, with the relations in its frontmatter. Everything on the other four tabs is a mapping out of this.

Full specification in docs/claim-format.md · the schema reference

A claim, and what a paper does with it

The format's central move is that these are two things. A claim is a proposition and is paper-independent; an assertion is one paper instantiating it, in a particular panel, with a particular analysis, taking a particular stance. The same proposition can be asserted by two papers that disagree about it.

The claim — paper-independent

uuid
— Canonical identity, generated once and never changed. It survives renaming and migration; the slug does not.
slug
— The human-readable filename, 3–5 words. The working identity, and replaceable — the UUID absorbs the identity.
doi
— Aspirational: the state in which claims are citable units. Empty for now, and never fabricated.
claim
— One declarative sentence, active voice, quantitative where the result is. If it needs two sentences it is probably two claims.
claim-type
— The epistemic character of the proposition. Seven values, below.
role
— What the claim does in this paper’s argument. Nine values, below.
addresses
— The research question this answers, as a q<N> id declared on the paper. Only a hypothesis and the alternatives competing with it carry one.
epistemic
— How strongly the claim is supported across its assertions.
concepts
— Controlled domain terms, for retrieval.
belongings
— Typed edges to other claims.

An assertion — one paper, one instantiation

paper-slug
— Which paper asserts it.
panel
— The figure panel, table or analysis block instantiating the claim here.
analysis
— The notebook or script that produces the result from data.
dataset
— The deposit used, with its DOI.
method
— The computational or experimental method.
confidence
— The paper’s own confidence, as distinct from ours.
stance
— What this paper does with the proposition: asserts, entertains, rejects, attributes.

The panel is a property of the assertion, not of the claim. That is the structural shift the format rests on: fig2a tells you how one paper chose to present a proposition, and the same proposition could be asserted in another paper's supplementary table, or with no figure at all.

Questions

A question is what the paper set out to answer, and it is deliberately not a node in the graph: a claim is a declarative sentence and a question is not. Each paper declares its own in index.md, and a hypothesis names one through addresses — as do the alternatives it competes with, which are the other answers to the same question. In the MIRA export they become real mira:Question nodes.

Reproductions

A third situation, beside the claim and the assertion: an agent re-enacting an assertion to see whether it holds. Each record says who ran it, when, against what, and what came out — and where it disagrees, the discrepancy is described precisely enough for the next person to diagnose it. “Numbers differ” is not sufficient.

A reproduction record is a fact about a run, not a verdict on the claim. What the site shows for each one, and why it is never phrased as verified, is on any paper page.

Seven claim-types

The nature of the proposition, independent of any paper. Distinct from role, which is what the proposition does in one paper's argument.

empirical A directly observed result from data.
interpretive An inference drawn from one or more empirical claims.
existence An assertion that a phenomenon or entity exists.
synthesis A claim integrating results across papers or datasets.
assessment A methodological or quality claim.
hypothesis A proposition bet on, not yet evidenced by this paper’s results.
prediction A deduced expectation, to be tested by an empirical claim.

Nine roles

Functional, not contentful: the same proposition can be empirical in one paper and control in another. The role is what the claim does here.

hypothesis A proposition the paper sets out to test.
prediction What the hypothesis entails, if it holds.
empirical A result the paper reports, anchored to a panel.
control A result whose work is to eliminate an alternative.
scope A boundary condition on where another claim holds.
methodological A capability that makes a downstream result interpretable.
synthesis A claim assembled from several others.
interpretation A reading of a result through a theoretical lens.
literature-context A claim from another paper, cited by DOI.

Relations 17 of 21 in use

Called belongings in the specification: typed directed edges between claim entities, and statements about logical structure rather than citations. When a claim requires another, it asserts that its validity depends on the other's — so invalidity propagates, and the graph makes that auditable. Counted from the corpus, not declared here.

supports 148 scopes 144 requires 104 enables-method 76 tests 75 confirms 70 dissociates-with 62 entails 54 derived-from 48 validates 44 interprets 36 extends 26 rules-out 22 part-of 4 in-tension-with 1 qualifies 1 refutes 1

Defined and never used: contradictsopposespredictsreplicates — which is a fact about the vocabulary worth settling rather than leaving. zmainen/elife-claim-trees#20

The specification is behind the corpus here. docs/claim-format.md §4 names five belongings — supports, requires, contradicts, extends, part-of — and the corpus uses 17 of 21 the schema defines. The rest were added as the work needed them and the document was not brought along. The counts above are the corpus; the specification is the thing to fix.

What it carries that no target does

Verification. That a claim was re-run, by what code, against what deposited data, with what value beside the published one. This corpus holds 177 such records. None of the four targets has a node type for one.

Stance. That a paper considered and rejected a proposition. An alternative exported without it reads as something the paper asserts — which is the inversion of its meaning.