nucleation

ADR-043: A warrant field on CLAIM, saying how a claim is supported (demonstrated, empirical, interpretive) apart from whether the record accepts it and in whose voice, piloted before it is required

11 min read

Properties3
StatusActive
Version2
Tagsrecord

Graph View

Context

At A8 of the exchange of 2026-10-14 the assistant sorted what the exchange’s cases had shown: “The key distinction is between what we’ve demonstrated by construction, what we’ve made plausible through examples, and what still requires empirical evidence.” The owner, at U9: “that’s a useful distinction, do we have anything like this already segregating CLAIMs? if not, we should consider adding it.”

What a CLAIM carries now:

  • status, the record’s credence, in THEORY’s words.
  • role, its voice: thesis, counter, granted, disclaimed, supposed (ADR-032).
  • grounds and rests_on, what it is supported by, without saying what kind of support that is.
  • line, with formal-apparatus as a second line for general results (ADR-040). That marks some formal claims, by subject matter rather than by warrant: a formal-apparatus claim can be Proposed and conjectural.

CASE has standing (how much is fact) and kind (ADR-042), and ARG has form. Nothing on the claim says how the claim itself is supported. Three Active claims can be a theorem, a finding reported in five studies, and a reading the record finds persuasive, and they look the same.

The exchange produced its own examples. CLAIM-294 and CLAIM-302 are Active by construction. CLAIM-296 is Proposed and interpretive. CLAIM-297 is Proposed and would become a demonstrated existence result if its model were analysed.

Decision

  • An optional warrant field on CLAIM, from a closed vocabulary. A9’s five terms, narrowed:
    • demonstrated: follows from stated premises, definitions or a construction. It licenses the conclusion under its assumptions.
    • empirical: supported by observation or experiment, scoped to the evidence.
    • interpretive: supported by conceptual argument, analogy or synthesis, settled neither way.
    • A9’s stipulated is dropped, as A9 suspected it should be: “it describes how a proposition is being used, more than how it is justified. The existing role: supposed already covers much of that.”
    • A9’s unestablished is held open for the pilot. It may be interpretive with status Proposed, in which case it is a second copy of status.
  • Single-valued, attached to the claim as stated. A9: “Split claims whose formal and empirical scopes are materially different rather than adding a multi-valued field that silently conflates them.” That is ADR-040’s practice for formal results, applied to warrant.
  • Never inherited. A9: “warrant: demonstrated must not propagate automatically along supports or rests_on edges. A demonstrated premise can support a conjectural empirical conclusion.” A lint could later report a claim marked demonstrated that rests on one that is not.
  • Orthogonal to status. Proposed with demonstrated is coherent: the proof holds and its interpretation is disputed.
  • Piloted, not required. Classify the claims filed from the 2026-10-14 exchange and a heterogeneous sample of older ones (A9 named CLAIM-197, CLAIM-199 and CLAIM-250), see whether the terms divide them without strain, then decide whether to require the field. A18: “test the classification on a handful of heterogeneous claims before changing the schema.”

What warrant records

Added 2026-10-19, at the owner’s decision on the promotion audit’s batch 028 question. Two readings of the vocabulary were in use. The pilot took warrant to say how a claim is supported now. Batch 022 took it to say how the claim would have to be settled, and relabelled CLAIM-052 empirical on that ground. The owner chose the first reading:

  • empirical means the record holds observations or experiments that bear on the claim as stated: a read study, a case with data, a measurement.
  • A claim that would stand or fall on an observation nobody has made yet is interpretive while the support the record holds is argument, analogy or synthesis. Its defeat condition can still be an observation. That is what makes it testable, not what warrants it.
  • When the observation arrives, the warrant changes with the claim’s version, like any other judgement the record makes.

The rejected alternatives were to read warrant as the kind of evidence that would settle the claim, and to add a fourth value for untested empirical claims. Status already says a claim is unsettled, so a fourth value would repeat it.

Alternatives considered

  • Nothing new: let formal-apparatus and prose carry it. The line marks subject matter, not warrant, and prose is what DP-3 says gets followed probabilistically.
  • Warrant on ARG and grounds instead of the claim. A9’s own objection: one proposition may be proved in one setting and supported empirically in another. The record already holds ARG form for the inference. The field is about the claim as stated, and the remedy for mixed warrant is a split.
  • A confidence scale. That is status. A9: “This is deliberately not another confidence scale.”

Consequences

  • If adopted: a vocabulary claim-warrants in luria.yaml, an optional field on CLAIM, the template documenting it, and the pilot’s classifications.
  • The warrant of a claim is the record’s judgement and can be wrong in the same way status can. It is versioned like status.
  • The risk A18 named: warrant becomes “another label applied to a compound proposition whose mathematical and empirical parts should have been split instead.” The pilot should count splits it forces.

Pilot results

Accepted by the owner on 2026-10-15. The pilot classified the thirteen claims of the 2026-10-14 exchange, A9’s three older claims, and the twelve claims filed on 2026-10-15 from the LEGO port and the database readings (the superseded CLAIM-306 left out), and applied the field.

claimstatusrolewarrant
CLAIM-197Activethesisdemonstrated
CLAIM-199Proposedthesisinterpretive
CLAIM-250Proposedthesisinterpretive
CLAIM-292Activegranteddemonstrated
CLAIM-293Activegranteddemonstrated
CLAIM-294Activethesisdemonstrated
CLAIM-295Proposedthesisinterpretive
CLAIM-296Proposedthesisinterpretive
CLAIM-297Proposedthesisinterpretive
CLAIM-298Proposedthesisinterpretive, after the split
CLAIM-315Proposedthesisempirical, split from CLAIM-298
CLAIM-299Activethesisdemonstrated
CLAIM-300Activethesisdemonstrated
CLAIM-301Proposedthesisinterpretive
CLAIM-302Activethesisdemonstrated
CLAIM-303Proposedthesisinterpretive
CLAIM-304Activethesisdemonstrated
CLAIM-305Activethesisdemonstrated
CLAIM-307Activethesisdemonstrated
CLAIM-308Proposedcounterdemonstrated
CLAIM-309Activethesisdemonstrated
CLAIM-310Activethesisdemonstrated
CLAIM-311Activecounterdemonstrated
CLAIM-312Activethesisinterpretive
CLAIM-313Proposedcounterdemonstrated
CLAIM-314Proposedcounterinterpretive

What the pilot found:

  • The three terms divided the claims, with one split. CLAIM-298 was compound, as proposed: its psychological form is now CLAIM-315 (empirical), on which CLAIM-298 (interpretive) rests. No other claim forced a split.
  • unestablished is dropped. Every claim it might have marked was interpretive and Proposed, so it would have been a second copy of status.
  • Conditions belong in the claim, not the warrant. CLAIM-302 and CLAIM-304 are demonstrated, conditional on pseudorandom generators; the condition is in their titles, and the warrant is plain demonstrated.
  • Citing a theorem does not make a claim demonstrated. CLAIM-199 rests on Blackwell’s theorem but concludes a reconciliation of two standards, so it is interpretive. The same holds for CLAIM-314, which applies an NP-completeness result for conjunctive queries to filed arguments in prose: the result is demonstrated, its application is not. Neither forced a split, because the theorem is cited from a THEORY rather than claimed.
  • Demonstrated within a model is still demonstrated. CLAIM-307, CLAIM-309 and CLAIM-310 follow from stated models. Whether the models fit people or bricks is what their “What it does not say” sections are for.
  • Orthogonal to status, as intended. Proposed and demonstrated occurs (CLAIM-308, CLAIM-313: the arithmetic holds and the reading is open), as does Active and interpretive (CLAIM-312).

Requiring the field. The owner, on the first version of this PR, which left it optional: “There’s no point adding this if we don’t use it”, and “‘failing’ is a forcing function to make you fill in the field. this is a feature, not a bug.” The field is required on every CLAIM, whatever its status, and every claim in the record was classified in the same contribution. The lint A9 suggested, a demonstrated claim resting on one that is not, is not built: that pattern is legitimate when the premise is stated as a condition, so it would be a report to read, not a violation.

Classification of the whole record

With the field required, the remaining 289 claims were classified the same day by eight readers working from this ADR, each reading every claim in its batch in full, and the results were applied without exception. With the pilot, every CLAIM in the record now carries a warrant: 87 demonstrated, 12 empirical, 216 interpretive.

Rules the full pass settled, beside the pilot’s:

  • An empirical hypothesis with no observation under it is interpretive. Readers disagreed here (CLAIM-105, CLAIM-220, CLAIM-263 and CLAIM-274 were first marked empirical); they were set to interpretive, following the pilot’s finding that unestablished would only repeat status. empirical means some observation the record holds supports the claim as stated, even if only a census of the record itself or a documented case.
  • A claim checked against the record’s own wording and definitions is demonstrated when the check is mechanical (CLAIM-200, CLAIM-210, CLAIM-211), and interpretive when it turns on how a thesis is read.
  • A supposition (role supposed) is interpretive. None of the three terms describes a premise taken for the argument’s sake; interpretive is the least wrong, and the role already says what kind of thing it is.
  • Superseded and Rejected claims carry the warrant of the claim as it was stated.

Compound claims. The readers flagged nineteen claims whose parts are warranted differently. Each was given the warrant of its main conclusion so that the field could be required at once; the rule above says they should be split. Four are already split in effect, by a separate claim holding the formal part. The other fifteen are left for the owner to split or keep:

claimwarrant giventhe split
CLAIM-056demonstratedLipschitz error recursion within a stated model; split: recursion demonstrated (cf. CLAIM-267) vs fit to real reconstruction chains (defeated_if) empirical, untested. Already split in effect: its model result is filed as CLAIM-267.
CLAIM-059demonstratedSplit: regime taxonomy and TV contraction bound are demonstrated; that real chains contract toward conventions is an empirical, untested hypothesis.
CLAIM-061interpretiveSplit: sufficiency half shown by constructed/existence cases (demonstrated); necessity half rests on a contested documented case and argument (interpretive). Already split in effect: CLAIM-229 already separates its two halves.
CLAIM-087interpretiveImprovement via image is a hypothetical example (interpretive; empirical test proposed); ideal-receiver bound is demonstrated (data-processing/garbling).
CLAIM-103interpretiveNesting of symmetries in transports is demonstrated (stochastic invertibles are permutations); their main use as yardstick is interpretive.
CLAIM-105interpretivePossibility is demonstrated by a constructed pair (CASE-037); that reconstruction chains do this is empirical, untested per defeated_if.
CLAIM-117interpretiveMain conclusion (fidelity requires preserving dynamics) is untested; split: commutator-defect identity demonstrated (algebra) vs fidelity relevance interpretive/empirical.
CLAIM-123demonstratedMain part: score addition composes only under conditional independence (derivation); split: binding failures with conditions met documented empirically.
CLAIM-142interpretiveMain: realist reading and evidential rule for probes; split: probe-relative equivalence (restricted Yoneda, density) demonstrated vs realist rule interpretive. Already split in effect: its general content is filed as CLAIM-264.
CLAIM-166empiricalMain microbiome conclusion rests on reviews (causal in rodents, modest in humans); time-indexed drive-profile part is interpretive.
CLAIM-178interpretiveNormative “manuscript should be a view” is interpretive; “drafts drop commitments and shift thesis silently” is empirical, from one exchange.
CLAIM-193demonstratedLinear-readout kernel result is proved (THEORYs 008/002/018/019, filed as CLAIM-269); carry-over to communicative realizations is interpretive. Already split in effect: its kernel result is filed as CLAIM-269.
CLAIM-196interpretiveApplies sheaf gluing to presented selves; split: demonstrated no-overlap-glues (CLAIM-197) vs empirical audience segregation/context collapse (boyd).
CLAIM-201interpretiveSplit: demonstrated odd-cycle construction (needs contextuality); empirical absence of contextual data (LIT-264); interpretive proposal that change of cover is surviving form.
CLAIM-219empiricalSplit: empirical premise that read data show no contextuality (LIT-264, LIT-851) vs demonstrated dilemma with 2-epsilon overlap bound.
CLAIM-240interpretiveMain is a conceptual reconciliation with task-relativity; split: empirical claim that holistic judgements are a held-out-predictive weighting (the defeat condition) vs conceptual consistency/defeat-condition requirement.
CLAIM-258interpretiveSplit: worked case (non-garbling, Blackwell-incomparable) demonstrated; that this is the normal translation case is a reading of CLAIM-091/compensation.
CLAIM-259demonstratedSplit: properties of D+_Q (directed, Blackwell-exact, symmetrization) proved; that it should be the decision term is a proposal with an empirical defeater.
CLAIM-264demonstratedSplit: exact half (restricted Yoneda, Cramer-Wold, marginals) demonstrated; evidential half on independently chosen held-out probes interpretive.

The splits, done at the owner’s request in a follow-up contribution. Of the fifteen, fourteen were split and one kept whole:

Five titles changed because they led with the part split off (CLAIM-059, CLAIM-123, CLAIM-166, CLAIM-219, CLAIM-264), and three defeat conditions lost the disjunct that moved (CLAIM-123, CLAIM-166, CLAIM-264). Four warrants changed: CLAIM-059 and CLAIM-259 to interpretive, CLAIM-219 to demonstrated (its dilemma, conditional on the split-off premise), and CLAIM-264 stays demonstrated with its evidential rule moved out. Each split original carries a dated note saying what moved where.

Backlinks