Editorial policy
Last updated
Sources and how we weigh them
Every review starts from a bounded set of public sources, each assigned a trust tier:
| Tier | Examples | Used for |
|---|---|---|
| A — primary | official App Store and Google Play listings, developer websites, patch notes, support pages | prices, purchases, ads, versions, dates, age ratings |
| B — independent editorial | established publications with dated articles | interpretation and analysis |
| C — community | forums and player discussions | recurring experience patterns, never sole proof of a fact |
| D — discovery only | search snippets, aggregators, reposts | finding sources, never for important claims |
When sources conflict we record the conflict, prefer current primary evidence for facts, and describe genuine disagreement on subjective questions. We respect robots.txt, never bypass paywalls or logins, and never copy articles: evidence is stored as short excerpts and every review is written as original synthesis.
Fact checking
Every factual sentence in a review is linked to the evidence that supports it. Before a review can reach a human editor it passes automated checks: numbers, prices, versions and dates must match the cited evidence; quotations must exist in sources; absolute claims need primary support; any statement implying first-hand play is rejected. A separate verification pass re-checks high-risk claims, and store metadata older than two weeks is refreshed before publication. Anything that fails is held for human review rather than published.
Scoring
Each AI critic scores the dimensions that matter to its persona (for example monetization fairness or touch controls) from 0 to 10. The overall critic score is a weighted average of the dimensions that the evidence actually covers, computed by code from published weights. If the evidence covers too little, the critic gives no score. There is no single combined site score.
| Score | Meaning |
|---|---|
| 9.0–10 | Exceptional within its target audience |
| 8.0–8.9 | Very strong |
| 7.0–7.9 | Good with meaningful caveats |
| 6.0–6.9 | Mixed, with notable weaknesses |
| 5.0–5.9 | Limited recommendation case |
| below 5 | Serious drawbacks for that critic’s priorities |
AI use and human oversight
AI models help extract evidence, draft the research dossier, write the critic perspectives and compose the article. They never browse freely, never publish on their own and never receive instructions from the pages they read. A human editor reviews every article against a written checklist before publication. Details: AI methodology.
Version and date sensitivity
Mobile games change constantly. Each review shows the reviewed version and the date sources were checked. Prices, purchases and ads are described as observed on that date.
Updates and corrections
Material changes — a new monetization model, a major update, a correction — are recorded in the review’s update history with the date. We never quietly rewrite a conclusion or present an updated article as newly published. See corrections.
Independence, advertising and affiliates
No developer, publisher or advertiser can pay for coverage, a score or a change to a review. Advertising, if shown, is placed by an ad network and is visually separate from editorial content; it never influences what we review or how. We do not currently use affiliate links; if that changes, affected links will be labelled.