1. Define one fair comparison
Each ranking must identify exactly what qualifies, who the comparison serves, and which product format, place, price range, and testing window apply. Different things do not get forced into one list for convenience.
2. Version the rubric before scoring
Every criterion needs a definition, evidence type, and weight. The weights must total 100 percent. A rubric receives a version and publication date before a governed review can use it.
3. Identify the exact item
A brand name alone is not enough. The record identifies the item, version, size, variety, lot, menu location, or other detail required to understand what was actually tested.
4. Record the test and evidence
- Preparation, serving method, location, date, and price when relevant.
- Repeat tests when a claim depends on consistency.
- Facts and measurements separated from sensory opinion.
- Known limits and unavailable comparisons stated directly.
5. Explain the score
Each criterion score needs a rationale connected to the rubric and evidence. Category scores are not automatically comparable across categories. Ties and no-score conclusions are allowed.
6. Disclose influence
Free samples, affiliate relationships, sponsored access, paid travel, ownership, employment, and other relevant ties must be recorded. Payment cannot determine eligibility, language, score, or rank.
7. Approve before publishing
A review must have an editorial approval and a separate fact-check approval. Draft, unverified, rejected, or incomplete records remain private and cannot enter public rankings.
8. Correct visibly
Material corrections, clarifications, retests, and retirements remain dated and attached to the affected review. A new version does not erase the earlier public record.
Automation and AI
Automation may assist normalization, drafting, accessibility text, code, and quality checks. It cannot claim to have tasted, visited, measured, witnessed, quoted, licensed, or verified anything. A human remains responsible for public facts, evidence rights, disclosures, scores, approvals, and corrections.
Community ratings
Public community ratings are not open. A future result must name its food target, eligible consenting-rater cohort, broad privacy-safe basis range, evidence state, snapshot date, exact-set release, method versions, and organic commercial status. The evidence label is threshold- and day-diversity-based, not a statistical margin of error. Exact participant counts and individual ratings stay private. Payment cannot change eligibility, weight, score, confidence, or publication, and a community rating never changes an editorial score.