Passing is not learning.

An online language platform teaches through theory videos of 5 to 15 minutes and assesses with multiple-choice questions. The videos are short, but holding attention through a traditional lecture format is hard. A missed detail means rewatching before each assessment, and a single lesson turns into hours. An analysis of millions of video sessions on edX reached something similar. Shorter videos hold attention better.

The alternative was to change the rules. With Claude, an AI assistant, the class vocabulary became a direct course, with points and a streak. Linking pinyin to meaning was enough to pass the platform. So three levels were added per topic. The first teaches with pinyin, audio, and short explanations. The second removes pinyin, gives fewer hints, and requires recognizing the characters. The third generates new sentences that mix vocabulary from earlier topics.

The result was telling. Modules passed on the platform did not hold up at levels two and three. Picking the right option out of four is recognition. Producing the answer without support is recall, and research on retrieval practice shows that this effort is what makes learning stick long term. Multiple choice measures the first and lets the second slip by.

The same method now applies to French, studied in person alongside. Accounts, sync, and more robust data structures would make the system sturdier, but they would add a barrier too early. The methodology gets validated first. What is missing gets built after. The course is published as Diǎn Diǎn.

Everyone learns differently, and every platform measures in its own way. It is worth confirming one's real level in detail, with a harder test than the one passed, because what each assessment measures can vary.