Our evidence standard
We separate four things that publishing arguments routinely blur: fact, estimate, inference and opinion. Every briefing names its sources, update date, confidence level and limits. Where credible organisations disagree, we show the disagreement instead of manufacturing certainty.
Source hierarchy
- Contracts, laws, filings, datasets and primary documents
- Professional bodies and methodologically disclosed surveys
- Named expert analysis and attributable reporting
- Anecdotes, used only to form questions—not to establish prevalence
Our five-part model
We analyse publishing routes through money, rights, work, reach and risk. No route receives an automatic moral score. Claims about sales, distribution or return on investment require definitions and comparable evidence.
Verification, including the machine kind
Every briefing is written and edited by a human being, then put through an AI-assisted verification pass: figures re-read against the source document, quotations checked against the original, links tested, claims flagged when the ledger does not support them. Technology is used to catch mistakes, never to decide what this site argues, and never to establish an uncited conclusion. Volume is deliberately low, because verification — not drafting — is the bottleneck, and because Google’s own spam policies are right that scaled content without original value deserves to rank nowhere.
Corrections
Material corrections are dated and described. Small spelling or formatting changes may be made silently. Sources can be challenged through the corrections desk.