Editorial policy

Where topics come from

We look for questions that get asked over and over and answered badly. The standard example is "my identity verification keeps failing" — most of the results repeat the same useless advice (upload your ID again), and almost nobody points out that the variable worth changing is usually the device and the lighting, not the document.

There is one hard rule on topic selection: it has to be a real operational problem, not a topic that happens to hold keywords. We do not write "what is blockchain" explainers, because nobody arrives at one with an actual problem.

How facts get checked

Content falls into three buckets, handled differently:

• Protocol-level facts (how a chain produces blocks, how a token is issued) — relatively stable, safe to state, linked to authoritative technical documentation where it matters.
• Platform rules (fees, limits, verification requirements, product terms) — we never hard-code the specific numbers. We describe the structure and the range, and tell you to check your own account page. A "checked on" date means the last time we verified that section, not a guarantee that nothing has moved since.
• Operational experience (which step stalls, what order works better) — written as general description. We do not invent precise timestamps, fill prices, amounts or yields, and we do not claim evidence we cannot produce, such as "measured in a live test".

That last one deserves an expansion, because false precision is the most common form of content fraud. "Tested at 14:23, completed in 3 minutes 17 seconds" reads as credible — and if it is made up, it is fabricated evidence. We would rather be vague and true than precise and false.

How far AI tools are involved

Plainly: AI assistance is used in drafting and organising the content on this site.

We think saying so is better than hiding it, and it is worth being specific about where it stops:

• Topics and structure are decided by people. What gets written, why, and in what order is not a model's call.
• Factual claims are not taken from model output. Anything touching platform rules, numbers or on-chain parameters is either sourced to an official page with a check date, or rewritten so it does not depend on a specific value. What a model says does not count as a source.
• No fabricated experiential evidence. No invented account screenshots, trade records, precise timestamps, or any first-hand material we cannot produce.
• Human review before publishing. Including whether internal links resolve, whether the risk warnings are actually there, and whether any absolute claims slipped in.

This does not guarantee zero errors. It does keep out the most dangerous category — invented numbers and fake evidence. Where we fall short of it, tell us.

The English edition specifically

The English pages are not machine translations of the Chinese ones. They were written against the same research, but the practical answer often differs by region: the payment rails are different, the regional restrictions are different, and "getting money out" does not mean the same thing for a reader in Kuala Lumpur as for one in Taipei. Where a section would have been wrong for English readers, it was rewritten rather than translated.

One consequence is worth stating: the two editions can disagree on specifics, and when they do, neither is the master version. If you read both and find a contradiction that region does not explain, that is a bug — tell us and it gets logged.

What we do not publish

• No coin, project or yield-product recommendations
• No price predictions or market views
• No "guaranteed", "risk-free", "capital protected" or "just park it there" phrasing — a return is never described without its risk
• No account profit, position or yield screenshots
• No third-party ads, sponsored posts or paid link placements
• No softening of a platform's risks because of commercial income

That last one is the hardest to hold yourself to, because our income does come from an exchange referral programme. Our answer is structural: keep almost every page free of referral links, and put no conversion path at all on the risk pages. The specifics are on the disclosure page.

Updates and corrections

Platform rules change and content goes stale. So:

• Errors get fixed and logged publicly. Every substantive correction goes in the corrections log — what changed and why. No silent edits.
• The "updated" date follows real changes. If the text did not move, the date does not move. We do not bulk-refresh dates to look fresh.
• Reader corrections come first. Point at a specific sentence and say what is wrong with it — that is the most useful mail we get.

Corrections go through the contact page.