How AAI Kernel research is published
An original AAI Kernel publication requires an answerable research question, retained evidence, a documented method, supported findings, disclosure permissions and accountable review. Publication makes a bounded result available for scrutiny; it does not make the result universal or confer external peer review. A promising question, an internal analysis or a publication-shaped draft remains unpublished research until the evidence and release conditions are satisfied.
Before publication: establish what was actually studied
The publication standard requires a corpus manifest connecting the research question to eligible captures, observation dates, systems, interfaces, languages, locations and retrieval conditions. Specify entity and source-role rules, exclusions, missingness and denominators before interpreting the result. The study must distinguish captured output from later source inspection and analyst classification.
Validate attributions that matter to the conclusion. Reproduce the calculations from identified inputs and retain the derivation version. Review whether contrary cases, sensitivity to reasonable classification choices and evidence gaps alter the claim. A reviewer should assess the result against its question, rather than accepting it because a table was generated.
Confirm permission for every public excerpt, image, source object and named comparison. A private commission does not become an original public publication by being selected for an article. An internal reviewer is not evidence of external peer review; describe the review actually performed.
Measurement and uncertainty explains what an eligible population permits a number to establish. The evidence standard keeps that result separate from its explanation and any proposed action.
What accompanies the article
A publication should present a direct finding followed by the research question, evidence scope, methods and validation, verified results, interpretation and limitations. Tables and figures need units, denominators, observation periods, method versions and evidence references in their captions or adjacent text. The reader should be able to distinguish an observed result from an explanation.
Record verified authorship and publishing responsibility, relevant commissioning or funding relationships, material conflicts and the nature of review. Do not invent contributors, credentials or disclosures. Separate the observation period, initial publication date and dates of substantive revisions; each must come from the actual accepted record.
Give a citation form containing the publication title, responsible author or organization, canonical URL, publication date and version. Link the reusable reference explanations for background; do not convert a dated result into a second definition of the subject.
Describe public and restricted evidence honestly
Reproducibility has several practical levels: inspecting permitted excerpts, repeating published calculations from released inputs, or rerunning the original collection. A publication must say which level is actually possible. Publishing a formula alone does not make inaccessible data independently reproducible.
Proprietary query protocols, client material and restricted source artifacts stay outside the public article unless disclosure is specifically permitted. A minimized evidence package may support verification, but its omissions and remaining restrictions must be visible. An access request is subject to the specific object and intended use.
If a central result depends on evidence that cannot responsibly be disclosed or sufficiently checked, retain the publication hold. Do not substitute a generic method page, synthetic specimen or unverified aggregate as public proof. Synthetic examples can explain a method when explicitly labeled; they cannot verify an empirical result.
After publication: preserve a usable research record
Material corrections should state what changed, why it changed and which conclusions remain supported. Preserve version lineage and explain any withdrawal where the result is no longer valid. A refreshed study with different data or methods should be distinguishable from the historical study rather than replacing its scope unnoticed.
The challenge process gives readers a route to identify a disputed claim and supporting evidence. Related summaries, downloadable objects and citations need review when their basis changes. The editorial requirement does not imply an implemented automatic correction service.
The publication collection may list only accepted articles with genuine abstracts, dates, methods and versions. Planned empirical pages remain absent from the public collection and route set. Publication is not a commercial endorsement of any organization described.
Definition
- Original research publication
- An original research publication is an AAI-authored research object with its own question, evidence scope, method, verified findings, limitations, disclosure permissions and accountable review and version record. A planned topic or internal analysis is not a published result, and internal review does not imply external peer review.
Common questions
- Is every article peer reviewed?
- No blanket peer-review claim is made. Each publication must state its actual review process and accountable contributors.
- Can research be credible when some sources are restricted?
- It can be, but the access limits and actual level of independent checking must be explicit. Restricted inputs do not justify claiming full open reproducibility.
07 Where this leads
Publication requirements become concrete when they are attached to a claim an editor needs to check. The editorial investigation example shows how scope, contrary evidence and rights shape that work. For a new commission, the media research pathway separates AAI’s research contribution from the newsroom’s editorial decision and publishing responsibility.