Evidence before output
Our process is designed around observable interface examples, first-party documentation, research methods, accessibility standards, and measurable experiment design. We do not present a copied archive page, a merchant description, or an AI-generated draft as completed research.
- Name the uncertainty. Every pattern must answer a concrete visitor question rather than decorate a page.
- Inspect the evidence. We look for recency, specificity, provenance, representativeness, and whether a claim can be verified.
- Review the placement. The proof should appear near the decision it supports without blocking or pressuring the user.
- Check ethics and usability. Dark patterns, fabricated activity, inaccessible motion, and privacy-invasive collection are treated as failures.
- Define measurement. We identify a primary outcome and guardrail metrics instead of assuming every conversion lift is healthy.
Research, experience, and opinion
We distinguish direct observation from second-hand information. When we have not personally used or tested something, we say so through the wording and production notes. Editorial judgment is explained with criteria instead of being disguised as an objective fact.
Use of AI and automation
Automation may assist with source organization, transcription, structured comparisons, grammar, or an initial draft. It may not publish autonomously. A human editor is responsible for checking claims, links, dates, relevance, originality, and whether the page adds value beyond its sources. Materially AI-assisted pages are marked in their production notes.
Updates
Pages are reviewed when source facts change, a reader reports a material issue, or the topic becomes stale. The visible updated date must reflect a meaningful change to the main content, evidence, structured data, or useful links rather than an automatic date refresh.