Editorial methodology

From question to publication

We define the scope, locate relevant sources, corroborate central claims and review names, figures, dates and units before publication.

Evidence, context and uncertainty

We distinguish observations, models, interpretations and hypotheses. We explain material limitations and avoid presenting forecasts or preliminary results as settled facts.

Automation and publication controls

Automation may assist discovery, classification, translation and routine checks, but publication still has quality gates. A story must have traceable sources, coherent language versions, valid relationships and an approved main image; incomplete items remain pending for review instead of being presented as finished work.

Editorial lifecycle and responsibilities

Work begins with a defined question, audience and scope. The research record gathers the claims to verify, primary evidence, useful context and any relevant conflicts between sources. Drafting, fact-checking, entity linking, translation, image selection and technical validation are separate stages, even when automation helps perform them.

Before release, the workflow checks that the headline is supported by the body, dates and units agree, sources are attached to the claims they support, related missions or objects are identified, and every public language version is complete. Publication is a decision point, not the automatic result of generating a draft.

Publication thresholds and uncertainty

Central claims require evidence proportionate to their importance. We prefer a primary record and seek independent corroboration when an assertion is disputed, consequential or based on a single interested party. Numbers are checked with their units, reference frame, date and stated margin of uncertainty; estimates are not silently converted into exact values.

Breaking information may be published only when its provisional nature is clear. If the available evidence cannot support a reliable headline, the item remains under review. Confidence labels and cautious language describe the state of knowledge; they are not used to disguise a missing source or an unresolved contradiction.

Post-publication monitoring and quality review

Publication does not close the editorial process. We monitor important source updates, mission schedule changes, newly released data and corrections issued by the institutions cited. Content linked to the same object or mission is reviewed as a set so that a changed fact does not remain correct in one page and obsolete in another. Reviews may lead to a correction, a dated update, additional context or withdrawal when the evidential basis no longer holds.

Quality is assessed through verifiable signals: source completeness, strength of primary evidence, language coverage, valid entity relationships, media provenance and unresolved review warnings. A score can prioritize work, but it cannot replace editorial judgement. Recurring failures become rules, tests or workflow changes so the system learns from errors rather than treating each incident as isolated.