Editorial and verification standards
A guide should make its basis for trust clear. The title area shows a concise verification label, review date and scope. The method, evidence and untested details live together in “Sources and verification notes” at the end.
How AI helps
Core Atlas uses AI assistance to find public references, compare announcements, organise steps, draft Chinese and English copy and check links. Generated prose, model memory and plausible inference are not gameplay evidence. The first batch was researched, organised and drafted with AI assistance. The counts below and each guide’s final notes identify whether independent gameplay testing exists.
We do not invent player identities, author biographies, playtime or testing experience, or generate images that pretend to document a test. Existing cave illustrations are original decorations, not screenshots.
Three verification states
- Game-tested (tested): actual tester, build, platform, conditions, actions and observations are recorded. Tests cover the critical facts, evidence files exist with matching checksums, and a human confirmation record is required.
- Source-verified (source_verified): key facts were compared against distinct source URLs with a recorded explanation of their support. There was no independent gameplay test. Wiki entries may share their sources and are not independent experiments.
- Needs review (unverified): evidence is insufficient, a material dispute remains or a version change requires review. This blocks production publication; changing a label does not upgrade the evidence.
What has been tested
Research records, counted once per topic: 0 game-tested, 6 source-verified, 0 needing review. Chinese and English share each record and are not counted twice.
No current topic has been game-tested. Tested guides must list the build, conditions, claims covered and human confirmation at the end; a documentary review will not be renamed a gameplay test.
How references are checked
Developer and publisher announcements take priority, followed by documented game data and tests, Wikis/databases, contextualised player reports and supplemental guides. Important claims need evidence links and a record explaining their support. Restricted access is disclosed, including when only indexed page text was available.
Conflicts lead to narrower scope, excluded unsupported numbers or a pending review. Player advice is not presented as an official mechanic. An announcement baseline identifies the information reviewed; it does not certify playtesting on that build. Automated checks catch missing metadata, files, broken references and inconsistent states; they cannot establish source accuracy or test authenticity on their own.
Incremental work starts by scanning existing topics and updating existing pages for the same intent. A separate review context reopens evidence for new research. A passed audit allows a preview; the owner still approves publication. Agreement between AI systems is not source evidence.
Version changes and publication
After an evidenced release is recorded, major changes flag affected topics as potentially outdated and add them to a recheck queue. They lose publication eligibility until their facts and evidence are reviewed. The site owner's publication approval remains a separate requirement.
Corrections
Please include the article URL, platform, build, world type, mods and reproducible observation with any correction. The owner can collect feedback through the project's review PR. Public corrections and copyright contact details have not yet been configured. Evidence and disputes are listed at the end of each guide; no fictional address or author identity is shown.