How Mandarin is actually learned
Long pieces on method: what tones really are, why word counts mislead, how review scheduling works, and what an examination does and does not measure. Written by the editorial team, reviewed before publication, and revised rather than quietly deleted.
从三级到四级：那堵墙
Everyone reports the same thing: the first levels came quickly and then progress slowed to a crawl. Nothing has gone wrong. The task changed underneath you, from recognising concrete words to choosing between abstract ones — and the methods that worked before stop working.
Yasmine Berrada · 2026年6月29日 · 阅读 6 分钟
HSK 之外：证书测不到的东西
HSK Guide’s editorial view, not a description of the examination’s design: a certificate measures a sample of a skill under standard conditions, and standard conditions are not living conditions. The gap we keep seeing, and what to do about it.
Yasmine Berrada · 2026年3月9日 · 阅读 5 分钟
听力为什么最难，怎么练
Reading waits for you; audio does not. Why comprehension by ear lags behind everything else in Chinese, how to tell a decoding problem from a vocabulary problem in two minutes, and a transcription-based protocol that actually moves the needle.
Nadia Haddad · 2026年2月9日 · 阅读 6 分钟
备考 HSK 到底需要多少词汇？
Vocabulary counts are the most quoted and least useful number in Chinese learning. Here is why raw totals mislead, what "knowing a word" actually means across recognition and production, and how to set targets you can act on.
Yasmine Berrada · 2025年11月17日 · 阅读 6 分钟
How these are written
The same editorial rules apply here as to the rest of the platform.
- Every article is versioned and attributed to a named author, and is reviewed by a second person before it is published.
- No official vocabulary list, syllabus, scoring rule or examination date is reproduced or invented. Where an authoritative figure is needed, the article says so and points at the examination authority.
- Chinese examples are tone-marked and checkable, and are held to the same standard as the dictionary.
- A model may help draft; a model cannot publish. Anything AI-generated is labelled and cannot leave the review queue on its own.
Found an error? 报告错误 — corrections are made against the article and its version history, not silently.