What is MuzuScore?
MuzuScore is Wanilog's community difficulty score: one number between 0 and 1 for every WaniKani item, built from real review results across the community. It exists because the obvious way to rank difficulty - accuracy - quietly gets the hardest kanji wrong.
Why accuracy gets it wrong
Take 暦, a level 45 kanji. The only people who review it made it through forty-four levels first - strong learners with settled routines. Everyone who found kanji hard quit long before meeting it.
A level-4 kanji is reviewed by everyone: speedrunners, strugglers, people in a terrible first month. So the easy item can post the worse percentage. Sort by accuracy, and the truly hard kanji hide behind the strength of their reviewers.
MuzuScore removes that filter.
How does it work?
Every hour, the community's newest reviews stream into Wanilog's scoring engine. It reads outcomes: what got missed, by whom, and how far each item fell.
How learners do against their own record. Wanilog works out how well an item's learners should do, from their history, then checks what happened. Missing below your own standard means the item is hard - even when everyone reviewing it is good.
How many people it bites. An item that knocks half its learners back down the SRS ladder is a different animal from one that torments a stubborn few.
How far the falls go. Losing an item from Enlightened costs months; a slip at Apprentice costs a day. Deeper falls count harder.
The model is proprietary to Wanilog, tuned against the whole dataset and re-derived as it grows. Every idea is on this page; the exact recipe stays private - the FAQ says why.
Reading the number
Most items sit low. Something near 0.15 gets cleared without drama.
An item up at 0.90 is a certified menace: learners miss it well below their usual standard, and the falls are the expensive kind. If an item you keep missing wears a deep red pill, the problem is probably the item rather than you.
No score yet? An item needs review history from at least 5 learners first. Under that, a score would mostly describe one or two people, so Wanilog shows nothing.
Hardest and easiest, right now
The ten hardest measured kanji next to the ten easiest - live, on the same scale, from the last hourly rebuild.
Hardest right now
| # | Kanji | Meaning | MuzuScore |
|---|---|---|---|
| 1 | 模 | Imitation | 0.96 |
| 2 | 授 | Instruct | 0.91 |
| 3 | 慣 | Accustomed | 0.91 |
| 4 | 識 | Discerning | 0.88 |
| 5 | 逮 | Apprehend | 0.88 |
| 6 | 徴 | Indication | 0.86 |
| 7 | 岐 | Branch Off | 0.85 |
| 8 | 還 | Send Back | 0.85 |
| 9 | 考 | Think | 0.82 |
| 10 | 敗 | Failure | 0.82 |
Easiest right now
| # | Kanji | Meaning | MuzuScore |
|---|---|---|---|
| 1 | 五 | Five | 0.00 |
| 2 | 犬 | Dog | 0.00 |
| 3 | 八 | Eight | 0.00 |
| 4 | 十 | Ten | 0.00 |
| 5 | 本 | Book | 0.00 |
| 6 | 父 | Father | 0.00 |
| 7 | 年 | Year | 0.00 |
| 8 | 気 | Energy | 0.00 |
| 9 | 何 | What | 0.00 |
| 10 | 皿 | Plate | 0.00 |
The full ranking is at /hardest-kanji. In the app, the score sits on item pages and next to your leeches.
Scores rebuild every hour from review results Wanilog users choose to share; the privacy page has the details. As for the name: ムズい (muzui) is Japanese slang for difficult.
FAQ
Why not just rank kanji by accuracy?
Accuracy on WaniKani is filtered by who survives long enough to meet each item. A level-50 kanji is reviewed only by learners who cleared the forty-nine levels before it, so its raw percentage reflects a strong, self-selected group and comes out flattering. MuzuScore compares each item against what its own learners' records predict, which removes the filter.
Why does one of my leeches show no score?
Items are scored once at least 5 learners have review history on them. Below that floor an aggregate would mostly describe one or two individuals, so Wanilog shows no score at all. Your private leech list still tracks the item; only the community number waits.
Does a high score mean I will struggle with an item?
Not necessarily. The score says the community as a whole found the item harder than their records predicted; your own history with it can run better or worse. The useful signal is agreement: when an item sits on your leech list and high on the community scale at the same time, the struggle is structural rather than personal.
Is the formula public?
No. Every idea in the model is described on this page, and in words the description is complete. The constants that tune it are not published: they are re-derived as the dataset grows, so a published formula would be stale within months and would freeze a model that is meant to keep moving.
How do I opt out?
Press the button below, or use the same control in Settings under Advanced. Your reviews leave the published scores on the next hourly rebuild, and nothing is deleted - your own stats and history are untouched, and opting back in is one press.
Checking...
How fresh is the data?
The aggregate is rebuilt every hour, and this page and the hardest-kanji ranking refresh on the same schedule. Individual scores move slowly, because each new review lands in a large and growing pool; expect the top of the ranking to reshuffle over weeks rather than hours.