Difficulty Scores

A Progressive Curriculum

The difficulty scale starts at 1, a rank held by exactly one character: 对 (duì, "correct"). That's a deliberate choice, twice over: 对 is one of the most useful words a beginner can own (it's how you agree, confirm, and understand your teacher's feedback), and it's the word our app is named after! From there the scale climbs through thousands of levels. You don't need to follow the scale rigidly, but it provides a reliable sense of progression. If you're comfortable with difficulty-50 characters, you know roughly what's coming at difficulty 60, and you can trust that the app won't throw difficulty-500 material at you before you're ready.

Curated Foundations

Every character's difficulty score is individually curated, and none of those judgments is ad hoc: we developed a written scoring rubric that lays out the system's foundational principles, described below, and every score is measured against it.

The Scoring Rubric

We developed the rubric to settle the rules before any scoring began, and it governs every placement in the corpus (how we apply it at that scale is a story of its own, covered on Built with AI). Its central question, asked of every single placement, is "what does learning this character let the learner do that they couldn't do before?" A character earns an early spot by what it unlocks: the compounds and conversations that become possible once you know it, given everything you've already learned. A few of its principles:

• A ladder, not a frequency table. In the rubric's words, "corpus frequency is not the same as pedagogical priority": corpus counts skew toward formal, written Chinese, but learners need conversational vocabulary first.
• Concrete before abstract. Things you can see, touch, taste, and do come before concepts, formal register, and literary particles, even when the abstract character is technically more frequent.
• Clusters open contexts. Characters that combine well are placed near each other, so learning a small group together opens a whole real-world context, like ordering in a restaurant.
• Learning tools come first. Characters that power teacher-student interaction, like 对 (duì, "correct"), 错 (cuò, "wrong"), 问 (wèn, "ask"), and 答 (dá, "answer"), earn early spots because they enable the learning process itself.
• Overrides are exceptions. For compound words, the computed score rules by default; "an override is an exception, not the norm."

The rubric is a living document, and so are the scores: refinement is iterative, with each pass nudging misplaced items toward their natural level.

Compositional Difficulty

For multi-character words and phrases and sentences, difficulty is calculated mathematically from the difficulty of the components. A word made of two easy characters is easier than a word made of two hard characters, but always harder than either character alone, because there's additional knowledge in how they combine. This means the difficulty scale is internally consistent: you can compare any two entries in the dictionary and know which one is harder.

Corrected Where the Language Isn't Compositional

Most Chinese words really are the sum of their parts, and for those the computed score is exactly right. But not all of them. Some compounds are much harder than their characters suggest: 个子 (gèzi, "height, stature") is built from "unit" and "child," and knowing both characters doesn't help you guess it. Others combine beginner characters into things a beginner has no use for, like 奶水 (nǎishuǐ, breast milk) from "milk" and "water." And a few work the other way around: you need 学生 (xuéshēng, student) or 咖啡 (kāfēi, coffee) in your first weeks, long before you'd study their component characters, so you learn them as whole units.

For these cases, the computed score gets a curated override, judged against the same scoring rubric, whose iron rule for compounds is that "a word's difficulty reflects when the learner knows enough characters to read it." Overrides push a word later when its meaning is opaque, specialized, or literary, and pull it earlier for survival vocabulary and loanwords. Less than 20% of words and less than 4% of phrases carry one; everything else stands on the computed score. The result is a ladder that follows the compositional structure of Chinese wherever it holds, and follows a human judgment call wherever it doesn't. And it's worth saying plainly: this ladder is one possible pedagogy among many. We designed it, we think it's good, and we keep iterating on it, updating our data and techniques as we learn from how our users actually learn. If you want the full rules, the mathematics page covers the override system in detail.