The Association holds its research to a published standard.
The Association's standing rests on a single claim: that its research can be checked. Every claim says what kind of claim it is. Every hypothesis states what would prove it wrong. Every source carries a grade. Every model earns its standing by testing. Every prediction is registered before the fact and scored after. The standard below is published in full, applied without exception, and versioned under founder change control.
The standard is published in full.
This page states the standard legibly; the documents govern. All three are public.
CRA Methodology Standards
The standard itself: classification, falsification, sources, models, machine assistance, framework representation, the two shelves, review, change control.
Start here. This is the document to read if you want the rules.
CRA Operating Procedures, SOP-1 to SOP-6
Who does what, in what order, on what timeline: publication, member submissions, source grading, forward claims, model review, Collection curation.
For researchers and reviewers: the procedures that put the Standards into practice.
CRA Contributor Field Guide
The standard in plain language, with worked examples and every template: claims tables, model cards, disclosures, registration forms.
For anyone who would rather meet the standard through examples than through the standard itself.
Every claim says what kind of claim it is.
Established Finding
Measured directly, or corroborated by independent high-grade sources, with nothing of equal grade contradicting it. Cited at the point of use and reproducible by a third party.
The excess length of day reached negative values repeatedly in 2024 and 2025 (EF; IERS C04).
Working Hypothesis
A proposed explanation consistent with the evidence but not established. Published only together with what it explains and the conditions that would falsify it.
The post-2015 tightening of the polar wobble reflects a change in core-mantle coupling state (WH).
Open Question
A material unknown, named explicitly. Where the evidence does not reach, the publication says so; suppressing an obvious open question is a standards violation.
Whether the current dipole decline continues past the 7,000-year minimum is an open question (OQ).
If [the observable] in [the instrument or dataset] shows [the threshold or signature] within [the window], the hypothesis is revised or withdrawn.
Conditions that no realistic observation could trigger do not satisfy the standard. When a condition triggers, the outcome publishes within one cycle, whether or not it favors prior work.
Evidence is graded, and the bar scales with the claim.
| Tier | Definition | EF | WH | OQ |
|---|---|---|---|---|
| S1 | Primary instrumental or archival data from an authoritative producer | |||
| S2 | Peer-reviewed literature, or independently replicated analysis | |||
| S3 | Documented secondary sources of identifiable provenance | |||
| S4 | Uncorroborated material |
Sources sharing a common origin count once: corroboration means independent derivation, not repetition. Evidence ages on a stated schedule, and every publication states the data-currency date of its key series. Where credible evidence of comparable grade points in opposing directions, both are presented side by side; nothing is silently reconciled or averaged away.
The plate is the ladder applied. Every event on it rests on a different quality of chronology, several are marked as debated, and the Association grades each at the point of use rather than grading the producers once and reusing the verdict.
Models earn their standing by testing, not by presentation.
Every model whose outputs appear in Association work carries a published model card: authors and declared track record, the assumptions register, input data with versions, calibration method, code availability, and its maturity grade. The declared history must be truthful; the gate on publication, however, is the validation record of the model, not the biography of the modeler.
Machine-assisted work meets the same standard as any other.
Material machine assistance is disclosed in the work.
A standard statement in the work identifies what was assisted: drafting, code, analysis, literature synthesis, or figure preparation. Routine tooling such as spell-checking needs no disclosure.
Machine output is never itself evidence.
Every quantitative result must re-derive from named data and stated methods by a third party without the assisting system. What cannot be re-derived does not enter the archive.
Sophistication is not evidence.
Review evaluates derivations, data, and reproducibility, and gives no weight to technical fluency. Polished packaging around an unvalidated model confers nothing.
The human author owns every claim.
Fabricated citations, invented data, or unverified machine-generated results under a contributor's name are misrepresentation, whatever produced them.
The Association applies this section to itself. Its research operations, engineering, and drafting make material use of machine assistance, operated under the rules above: machine output is never evidence in Association work, every quantitative result re-derives from named data and stated methods, and the named human authors own every claim.
Other researchers' work is represented faithfully or not at all.
The Association describes research by topic field. A researcher's named theory, framework, or brand appears in Association work only with written consent, and only where the work is faithful to its published specification. Attaching a framework's name to work that materially departs from it is a standards violation, in sympathetic and critical treatments alike.
| 01 | The claim is made | A work claims to model, test, extend, or refute a named third-party framework. |
| 02 | The originator reviews | Before the name is used, the framework's originator is offered a fidelity review against their published specification, with a stated response window. |
| 03 | Name, or rename | With consent and confirmed fidelity, the name is used. Without either, the work still publishes on its merits, renamed to describe what it actually models. |
Consent governs the use of the name, never the publication of the analysis. The full property terms, including the researcher-retains-everything partnership model, are in the Standards and the Research Partnership Standards.
What the Association reviewed, and what it curates, never blur.
Work reviewed under this standard: Association publications, accepted member submissions, prepared datasets, and model cards. Carries CRA identifiers and the standard version it was reviewed under. Versioned; history is never silently rewritten.
Curated external work held by link, attribution, and source grade: analyses, threads, talks, and documents that inform the field. Nothing is absorbed or reproduced; owners direct their entries and may remove them at any time. Curation asserts relevance, not review.
Presence in the Collection confers no elevation; the only path between shelves is review. Commissioned work is paid under the partnership honoraria schedule, and a finder's honorarium recognizes members who surface material that becomes commissioned work.
Every prediction is registered before the fact and scored after it.
Every submission travels the same path into the archive.
Every Association publication passes six audits before release: labels, falsification, sources, figures, models, and framework names. Member submissions travel the same standard through a structured review; the portal opens with the platform. Acceptance asserts conformance with the standard, never endorsement of every conclusion, and attribution always remains with the author.
| 01 | Intake | Structured submission: metadata, methodology declaration, datasets and versions, model cards, disclosures, claims summary. |
| 02 | Screen | Scope and facial conformance, including the framework-name check. About ten business days, with structured feedback on failure. |
| 03 | Peer review | Two reviewers apply the audits; model-bearing work routes through validation review in parallel. About thirty days. |
| 04 | Resolution | Accept, revise and resubmit, or decline, always with written rationale. Declining implies no judgment beyond the standard. |
| 05 | Publication | Into the versioned archive under the author's name, with the standard version recorded. Attribution stays with the author. |
The Field Guide sets out the standard as checklists that take an afternoon.
The Contributor Field Guide restates everything on this page as working checklists and templates: claims tables, model cards, disclosures, and the submission checklist.