Surviving WaniKani's Death and Hell levels
The names are WaniKani's own. Any level page sorts the sixty levels into six labeled bands: 快 Pleasant (1-10), 苦 Painful (11-20), 死 Death (21-30), 地獄 Hell (31-40), 天国 Paradise (41-50), and 現実 Reality (51-60). They are in the URLs too: ask for kanji with difficulty=paradise and WaniKani knows what you mean. Asked twice on the forum what the names mean, WaniKani has never answered; the closest thing to a reply is that they are flavour text.
The folklore that grew around them says Death is where WaniKani loses its users and Hell finishes the job. The data says something smaller, stranger, and easier to plan around.
Nobody quits here
Three community scrapes have mapped where users actually sit, and none found a cliff at level 21. A 66,628-user pull in 2019 came out as a smooth exponential decay: steepest across the free levels, flattening after, with a pile-up at 60. A 2017 snapshot counted 7,849 accounts at level 1 and 1,444 still there by level 5. The largest rerun, 109,468 accounts in 2022, found 2,211 at level 60.
By level 21 the quitting is mostly over. All three scrapes count forum-linked accounts, so real early attrition is presumably steeper still. Whatever the Death levels do, they do it to the people who stayed.
Six months in, the burns come back
Take the first item you ever learned. At perfect accuracy it reaches Guru I in 37 hours, then waits 7 days for Guru II, 14 for Master, 30 for Enlightened, and 120 more before its burn review. Your oldest items come back for their final reviews just under six months after you started, with everything learned since queued behind them.
That is the wave, and it runs on time rather than level. At full speed six months lands you at level 20-21, exactly where the band begins. At ten lessons a day it finds you in the teens, in what the site calls Painful. Same wave, different postcode.
It is also smaller than the folklore says. One learner who measured it across the arrival watched a 150-review daily average become 175: a bump you feel, not a doubling. What sets the plateau afterwards is lesson pace, not the band you are standing in.
Reported daily review loads
One row per learner's reported figure. The two measured averages bracket the burn wave's arrival around day 180; the ranges are plateau reports at speed. No dataset shows a band-by-band spike pattern.
Heavier, not harder
The long-thread consensus is that the kanji themselves never get harder; what gets harder is maintaining an ever-growing total. There is real dissent. Some veterans rate the abstract kanji of levels 11-20 as the worst stretch of the whole journey, and the level-21 arrival who opened the survival thread found the new kanji noticeably more abstract than what came before. Treat "it never gets harder" as a majority opinion rather than a law.
Hell is not a second spike
A widely shared workload graph once showed load cresting in Death, dipping through Hell, and spiking again in Paradise. When the forum dug into it, the graph turned out to count items sitting in the SRS rather than reviews per day, which says nothing about what a given Tuesday feels like. The testimony does not line up either: one thread on Paradise has someone reporting the 40s as easier than the 30s while another, mid-Hell, is still underwater.
There is one genuine late rise, and it only reaches full-speed learners: the fast levels from 43 on compress the lesson schedule, so load climbs again at 41+. That has nothing to do with Enlightened items coming home. Through levels 31-50 your pace, your accuracy and your leech count set the load. The band name does not.
What survivors did
Most of what the survival thread's veterans recommend is not queue management. They read in the wild, because meeting a leech in a real sentence fixes it in a way another review round does not. They write similar kanji side by side to break lookalike pairs. They replace mnemonics that stopped sticking with their own. And they pause on purpose: one sat at level 20 for 50 days by choice and walked into Death with a clear queue.
For the queue itself there is one dial, and it is the Apprentice count. Lessons are the only input you control, so stop them when Apprentice climbs past your line and the queue drains itself within days. The lessons-per-day guide has the number and the arithmetic behind it.
Accuracy sets the floor under all of it. Sustained 85-95% is healthy; a long stretch in the 60s means the items are not known well enough to be climbing. Below roughly 67% the queue grows about as fast as you clear it, and no schedule saves you. The fix at that point is drilling the specific items that keep failing rather than grinding the whole queue harder.
You can watch your own wave approach. Wanilog projects the next 60 days of expected reviews from your current SRS state, so a climbing peak in that graph is the wave, weeks out. Connect your API key to see where yours is.
If the queue is already out of hand
Stop lessons entirely and let the Apprentice count fall back before restarting them; the queue stops growing the day the lessons do. The rest of the recovery playbook is in our burnout guide.