Loading this page…
The standards behind every piece we publish. Each clause is dated, and changes are logged like any correction.
Evidence first; context before conclusion; separate fact from interpretation; make methods visible; acknowledge uncertainty; correct errors in public. Programmes, topics and metrics are editable records, not hard-coded architecture.
Every quantitative claim that is marked major, and every chart, names a source type: official statistics, administrative record, peer-reviewed study, survey, media report, or ContextLens derived. We do not publish a single reliability score.
Definitions, units, geography, period and missing-data rules sit with each indicator and dataset. Missing observations are gaps, never zeros. Charts are pinned to the dataset version they were drawn from.
Every paper and dataset carries a status and a version number. The status tells you how settled the work is; the version tells you exactly which edition you are reading or citing.
Draft policy · for your review
ContextLens uses AI tools as research infrastructure, in the way a researcher uses a spreadsheet or a search engine. They help with the work; they do not decide what we conclude. We may use AI to find and organise documents and tables, help write and check data-processing code, flag inconsistencies between sources, and suggest edits for clarity. We never use AI to produce a finding, figure or citation without a person checking it against the source; never to generate sources, quotes or data; never to present an automated classification as fact (see clause 7).
A person accountable for the piece checks sources, figures and interpretation before publication. Reviewers are named on the piece when they reviewed it.
Where work involves judgement - framing or sentiment in media - we publish the coding scheme, who coded, and how often coders agreed. Negative coverage is not treated as bias by default. Automated labels are never presented as objective truth.
Cite codes never change. A code is CL-P, CL-R or CL-D, the year of publication, and four digits. Cite the code and version. Slugs may change.
Errors are dated and public. Revised work keeps a version history. Superseded work stays online so old citations still resolve.
Every major quantitative page states what the measure cannot tell you. Sequence is not treated as cause.
Terms used in indicators and papers are defined on those pages. A fuller glossary will grow with the corpus; unused terms are not listed.