A Daily Kanji Study Routine: How Many Kanji per Day, Review-Load Math, and the Three-Block Schedule
The right answer to "how many kanji per day" has two numbers and three blocks: a new-card rate, the steady-state review queue that rate produces, and a morning + commute + evening schedule that spreads out the work. The rest is arithmetic.12
Overview
A daily kanji study routine is not just a count. It is a small system: one knob (new cards per day) determines a much larger downstream cost (the standing wave of reviews that follows). A schedule then splits that cost into short, distributed sessions so retention survives.134
What a daily kanji routine actually is
A daily kanji routine combines three quantities: a new-items-per-day rate, the review queue that the SRS produces from that rate, and a within-day schedule that spreads the review work across multiple sessions. The first is the learner's lever; the second and third are the costs that follow from it.142
You set the new-card lever inside the SRS tool as a numeric cap. In Anki, "New cards/day" is the deck option that limits how many new cards the program can introduce each day it is opened.1 Amenokori's free tier fixes the cap at 20 new and 150 reviews per day.5 WaniKani delivers new items via its level-up gating rather than a learner-set cap.6
The review queue is not directly chosen. It is produced by SM-2-class (or FSRS-class) interval scheduling acting on accumulated cards. Under SM-2, intervals are I(1)=1 day, I(2)=6 days, and I(n)=round(I(n-1)·EF) for n>2, with initial easiness factor EF=2.5 (minimum 1.3).7 Every new card introduced today re-enters the queue at day 1, day 6, day ~14, day ~35, day ~90 and beyond, producing the review-load curve discussed below.
A within-day schedule that puts the work in multiple short sessions, rather than one long block, is itself evidence-based. Spacing is one of the four "desirable difficulties" Bjork and Bjork identify as supporting long-term retention.89 Cepeda et al.'s meta-analysis of 184 articles, 317 experiments, and 839 distributed-practice assessments documents distributed practice outperforming massed practice in the overwhelming majority of comparisons across domains, materials, and populations.4
Why "how many per day" is the wrong first question
Setting the new-card rate without sizing the review queue reverses the real cost structure. New cards are cheap (one short encoding event). Reviews are recurring (each card re-enters the queue many times across years). The Anki manual states the relationship directly: if you are consistently learning 20 new cards a day, daily reviews settle at roughly 200 cards per day.1
Industry summaries make the same point: "If you add 20 new cards a day, you might eventually face 200-300 reviews daily" and "every new card creates a trail of reviews for weeks to come."2
The right planning quantity is the steady-state review queue the learner can tolerate, not the exciting day-one new-card number. Reverse the question and the answer changes.
The two numbers and the three blocks
Two numbers: a new-items-per-day rate, and a maximum-reviews-per-day budget the learner is willing to spend at steady state. The Anki manual exposes both as deck options.1
Three blocks: a morning encoding block (new items), an opportunistic midday block (reviews in a 5 to 15 minute gap, usually a commute or break), and an evening cleanup block (remaining reviews plus the day's leech triage). Here, leech triage means identifying cards you keep failing and deciding whether to suspend, rewrite, or relearn them. The three-block frame is a direct application of the distributed-practice principle.349
The diagram above is the article's spine. The four H2 sections that follow justify one piece each: the rate ladder, the calendar timelines, the review-load curve, and the distribution schedule.
How many new kanji per day, by intensity
Four rates cover the realistic range for adult self-study: 5, 10, 15, and 20 new items per day. Each rate has a learner profile that fits it and a steady-state review cost.
| New cards/day | Time budget | Typical profile | Steady-state reviews/day under SM-212 |
|---|---|---|---|
| 5 | ~15 min total | Sub-30 min/day budget, parents, full-time workers, post-burnout restart | ~30 to 60 |
| 10 | ~30 to 45 min | The modal self-study cadence | ~80 to 120 at 90% accuracy; ~145 at 80% |
| 15 | ~45 to 75 min | Fixed exam date 12 to 18 months out | ~120 to 180 |
| 20 | ~75 to 100 min | 90+ min/day, no other major SRS deck | ~150 to 200 (up to ~300 at low accuracy)2 |
5 new items per day: the sustainable baseline
The 5-per-day rate is the floor this article recommends when the learner is unsure. It is much faster than the formal Japanese primary-school pace (the 学年別漢字配当表, the grade-by-grade kanji allocation table, introduces 80 kanji across Grade 1, about 5 per week of school, well under 1 per day equivalent). But it is still well under the rates that produce the steady-state queues most adult learners cannot sit with at week six.11
Industry summaries place this rate inside the conservative end of the recommended range: "5–10 new kanji; prefer fewer with higher accuracy."12
The 5/day rate is "sustainable" because it leaves substantial headroom for accuracy collapse, illness, travel, and life events. It is not a separately evidenced pedagogy; it is the rate that lets the routine survive bad weeks.
10 new items per day: the standard self-study rate
The 10-per-day rate is the modal recommendation in industry-grade guides.12 At the Anki-manual ratio (20 new produces ~200 reviews/day), halving the input roughly halves the steady-state count. That puts 10/day at about 100 reviews/day under SM-2 default ease at 90% accuracy.1
The calendar arithmetic to 1,000 kanji at 10/day is 100 days on paper. The realistic multiplier is discussed in the next H2.
15 new items per day: the JLPT-deadline rate
The 15/day rate is not quoted separately by the Anki manual or by the canonical industry guides. It sits between the 10/day self-study standard and the 20/day Anki-manual illustration. The numerical anchors in this band are interpolated from those two flanking rates.1212
This is the rate at which the steady-state review queue first crosses ~150 reviews/day, a common shoulder where self-studiers begin reporting review fatigue.
20 new items per day: the upper bound for most learners
20/day is the Anki manual's canonical illustration: "consistently learning 20 new cards a day…roughly about 200 cards/day."1 The same figure recurs in industry guides ("200-300 reviews daily").2
20/day is also the Amenokori free-tier ceiling on new cards. The corresponding 150 reviews/day cap is hit at this rate under steady-state SM-2 or FSRS scheduling.5
WaniKani delivers approximately 20 items/day on average over a 60-level path completed in 12 to 18 months under the standard apprentice timings (4 h → 8 h → 1 d → 2 d before reaching Guru 1).6 WaniKani items mix radicals, kanji, and vocabulary; the "20 items/day" figure is not "20 kanji/day."
Beyond 20: the diminishing-returns zone
No primary citation in this article supports the 25 to 50 per day band as a recommended pace. The Anki manual stops its worked example at 20.1 The Bjork and Bjork desirable-difficulties framework predicts that pushing the encoding rate past the point where reviews fail to clear shifts the routine from spaced practice to massed practice with relearning. The literature documents that shift as worse for long-term retention.49
Heisig-style sprint timelines for the meaning-only-first variant exist outside the cited sources. The per-day numbers in that variant are not directly comparable to readings-included pace.
What "an item" means depends on your method
The per-day number is not directly comparable across methods.
| Method | What counts as one "item" | Implication |
|---|---|---|
| WaniKani | Mixture of radicals (~5–7/level), kanji (~30/level), vocabulary (~100+/level)6 | 20 items/day averages across all three categories |
| Heisig RTK | One kanji-to-keyword pair | Meaning only; readings are a later layer |
| Anki vocab-first deck | One vocabulary word that may share kanji with others | 10 cards can teach fewer than 10 distinct kanji |
| Anki kanji-with-readings deck | One kanji with one or more readings | Each card is a multi-fact encoding event |
Recognition demand changes accordingly. Ten vocabulary cards that share kanji are not the same encoding work as ten isolated kanji.
Implied timelines to 1,000 and 2,136 kanji
The new-card rate converts directly into a calendar. The math has two layers: the paper math (rate × days = kanji) and a realistic multiplier that accounts for the drag every routine encounters.
The paper math vs. the real-world math
Paper math: 10/day × 100 days = 1,000 kanji; 10/day × 214 days ≈ 7 months for the full jōyō (2,136).13
The realistic calendar slips behind the paper calendar for three reasons. First, missed days (illness, travel, life events) delay new-card introduction without reducing the maintenance review load. Second, failed cards re-queued under SM-2 reset the repetition counter and interval to 1 day on grades q<3, extending the working time per kanji.7 Third, accuracy-driven slowdown, discussed in "Accuracy is the hidden variable" below, adds more drag.
A 1.3× multiplier is a planning convention this article uses to convert paper days to elapsed calendar weeks. It is not a measured value. The figure is an inference combining the Cepeda meta-analysis (distributed practice raises mean retention but does not eliminate forgetting) with the Lally habit-formation data (median 66 days to behavioural automaticity, range 18 to 254).414
To 1,000 kanji (roughly N3 recognition)
| New cards/day | Paper math | Realistic (1.3×) |
|---|---|---|
| 5 | 200 days (~6.5 months) | ~8.5 months |
| 10 | 100 days (~3.5 months) | ~4.5 months |
| 15 | 67 days (~2.5 months) | ~3 months |
| 20 | 50 days (~7 weeks) | ~2 months |
1,000 kanji corresponds roughly to N3 recognition territory. The pre-2010 JLPT 出題基準 (test content specifications) Level 2 target was about 1,000 kanji, and N3 sits between old Levels 3 and 2 in the post-2010 mapping. This milestone is most usefully framed by JLPT rank, since the recognition-vs-production distinction matters more for a learner's daily decisions than a single percentage.
To 2,136 kanji (the full jōyō target)
The jōyō ceiling, 2,136 characters, was established by 平成22年内閣告示第2号 (Cabinet Notice No. 2 of 2010, November 30, 2010) as "現代の国語を書き表す場合の漢字使用の目安" (a guide for kanji use in writing modern Japanese).13
| New cards/day | Paper math | Realistic (1.3×) |
|---|---|---|
| 5 | 427 days (~14 months) | ~18 months |
| 10 | 214 days (~7 months) | ~9 months |
| 15 | 143 days (~5 months) | ~6 months |
| 20 | 107 days (~3.5 months) | ~4.5 months |
Finishing the jōyō list is not the same as reading Japanese fluently. The jōyō set covers approximately 99% of newspaper kanji tokens (Bunkachō 2007 on the Asahi + Yomiuri 2006 sample). But residual proper nouns, place names, specialist vocabulary, and hyōgaiji (characters outside the jōyō list) remain outside it.15
Why most learners finish slower than the math says
Distributed practice is a robust effect, but it does not eliminate forgetting. Cepeda et al. (2008) document that the optimum inter-study gap is approximately 20% of the test delay for delays of a few weeks, falling to about 5% for one-year delays.3 A learner whose review intervals drift from this ridgeline (because of skipped days, queue debt, or accuracy collapse) loses retention and sends cards back into the early queue.
Habit-formation data put the median time to behavioural automaticity at 66 days with a range of 18 to 254 days across individuals.14 The routine does not become low-effort overnight.
SM-2 lapse behaviour compounds the drag. On a grade of q<3, the repetition counter resets to 0 and the interval returns to 1 day. The easiness factor decreases by approximately 0.2 per low grade, with a floor of 1.3.7 A single lapsed card therefore costs roughly one full re-encoding cycle. It also slows that card's interval growth until corrected.
The review-load curve every routine has to plan for
Reviews are not a one-time price for a card. They are paid many times across years, on a schedule the learner does not set. The shape of the curve is determined by SM-2 (or FSRS) interval geometry, not by the new-card cap; the cap only sets the amplitude.
The standing wave: new cards create reviews for weeks
Under SM-2's default interval sequence, a card introduced on day 0 generates a review on day 1, day 6, day ~14 (6 × 2.5), day ~35, day ~90, day ~225, day ~560 and beyond, assuming default EF=2.5 and correct grades q≥4.7 The early intervals dominate review load.
Sustained new-card input creates a superposition, with several review layers arriving on the same day. The day-60 queue contains day-1 reviews from yesterday's new cards, day-6 reviews from cards introduced six days ago, day-14 reviews from cards introduced ~14 days ago, day-35 reviews from cards introduced ~35 days ago, and a small tail of day-90+ reviews from earlier introductions.
This is the standing wave: a repeating review load created by earlier new-card days. The queue at day N is the sum of contributions from all preceding introduction days whose next review falls on day N. The cap sets the wave's amplitude.712
Where the curve peaks: weeks 4 to 6
The peak follows from the SM-2 interval geometry, not from observation. The first three intervals (day 1, day 6, day ~14) carry the heaviest review traffic per introduced card because they fire at the shortest spacing. The fourth interval (day ~35) is the first one outside the four-week window. So the peak instantaneous queue occurs when the day-35 interval starts firing for the earliest-introduced cards, around weeks 4 to 6 after launch.7
Concretely, cards introduced on day 1 fire their day-35 review at week 5. Cards introduced through day 7 fire their day-35 reviews across weeks 5 to 6. From week 6 onward, the day-1 + day-6 + day-14 + day-35 layers all contribute to the queue at the same time. The queue size from week 6 onward is the steady-state value rather than a growing curve.
The implication for the routine is the "do not raise your new-card rate in week 1" rule. The day-1 queue of a 20/day routine is 20 reviews, the day-7 queue is ~70, but the week-6 queue is the full steady-state ~200/day load.12 The week-1 experience can underestimate the cost by roughly an order of magnitude.
| Week of routine | Approximate queue at 20 new/day | Why |
|---|---|---|
| 1 | ~40 | Only day-1 layer firing |
| 2 | ~70 | day-1 + day-6 layers |
| 3 | ~100 | day-1 + day-6 + early day-14 |
| 4 | ~140 | day-14 layer at full strength |
| 5 to 6 | ~200 | day-35 layer fires; full steady state |
| 7+ | ~200 | Steady state; small day-90+ tail |
Steady-state review counts per new-card rate
The Anki manual's canonical anchor is the key number: 20 new cards/day produces ~200 reviews/day at steady state.1 The implied ratio is roughly 10 reviews per new card per day at the plateau, under SM-2 default ease and ~90% retention.
Linear scaling at the same ratio: 5/day → ~50 reviews/day; 10/day → ~100; 15/day → ~150; 20/day → ~200.1 These first-order estimates assume the same ease distribution and the same retention target.
Industry-summary anchors confirm the same shape. Nihongo Path says to "expect 60–120 [reviews/day] once momentum builds."12 This maps to a 6 to 12 new/day rate at the Anki-manual ratio. StudyCards AI says, "20 new cards a day, you might eventually face 200-300 reviews daily."2 The upper end (300) reflects accuracy below 90%.
Amenokori's free-tier review cap of 150/day corresponds to the 15 new cards/day equivalent under the same ratio.5 The free tier therefore supports up to the JLPT-deadline rate without paywall.
Why FSRS scheduling smooths the curve
FSRS models three per-card memory variables (difficulty, stability, retrievability) and chooses each card's next interval to land at a target retrievability set by the user (default 0.90 in Anki).16171 By contrast, SM-2 applies a global easiness-factor update that does not distinguish "this card has high stability after three correct grades" from "this card just came back from a lapse."7
The open-spaced-repetition benchmark, run across 9,999 Anki collections containing approximately 350 million filtered reviews, reports that FSRS produces more accurate recall predictions than SM-2 for 99.5% of users tested.10 More accurate prediction permits longer intervals at the same retention. In practice, that means 20 to 30% fewer reviews for the same retention target.10
At the same new-card rate and the same 90% retention target, an FSRS-scheduled deck plateaus at roughly 70 to 80% of the SM-2 review load. A 20/day routine that hits ~200 reviews/day under SM-2 hits roughly 140 to 160 reviews/day under FSRS at matched retention.110
Accuracy is the hidden variable
The Anki manual is explicit on the workload-versus-retention trade-off. At 0.90 desired retention, the interval will be 100 days; at 0.95, the interval falls to 46 days, which roughly doubles review frequency for the same set of cards.1
The reverse direction is the important failure mode. A routine that aimed for 90% retention but is delivering 80% retention has cards landing in the queue more often than the schedule budgeted. The 200 to 300 review range StudyCards AI quotes for 20/day is bounded above by this drift.2
The decision rule follows: if measured accuracy is slipping below 85%, the bottleneck is encoding, not review time. Adding review time at low accuracy worsens the standing wave. Reducing new cards is the corrective lever.110
Distributing reviews across morning, commute, and evening
The three-block schedule is not an aesthetic preference. It is a direct application of distributed practice: short sessions across the day outperform one long session at matched total minutes, by a margin that the Cepeda meta-analysis documents across hundreds of comparisons.349
Morning: the new-items block
New material is most fragile at the day-1 interval. It benefits most from spaced retrieval across the day, not from being encoded in the same session as the evening's reviews.34 Five to fifteen minutes, ideally before the day's first inbox check, is enough for the new-card budget at any of the four rates.
The Bjork and Bjork pedagogical re-statement of the desirable-difficulties framework lists spacing alongside interleaving, varying conditions, and retrieval testing as the four core mechanisms that appear to slow apparent learning but enhance long-term retention.9
The morning block is the encoding session. The evening block is the first long-interval recall. For a one-day retention interval, Cepeda et al. 2008 place the optimum gap in the range of hours, not minutes.3
The rule: never start the day with reviews from yesterday's backlog. Backlog goes in the evening block.
Commute (or any 5 to 15 min gap): the review block
Short opportunistic blocks fit reviews because a review is a recognition task. The prompt is on the screen, the learner retrieves the answer, and the system grades it. The encoding effort of a new card (read explanation, generate or check the mnemonic, verify reading) does not fit a 5 to 15 minute interrupted window in the same way.
Distributed-practice meta-analysis result: the gain from spacing is preserved across a wide range of within-day gap sizes, in the overwhelming majority of comparisons.4 A commute-sized 10-minute block can contribute the same per-minute retention value as the evening block.
The rule: aim to clear half the day's review queue in this block.
Evening: the cleanup block
The evening block addresses the residual queue: cards that hit their review interval after the commute block, cards that lapsed during the day and were re-queued under SM-2's "back to 1 day" rule7, and the day's failed cards that need a second pass.
The evening block enforces queue-zero at sleep. Under SM-2's interval geometry, the standing-wave amplitude is the same either way, but overflow delays its arrival time. Tomorrow starts in deficit if today ends with a non-empty queue.712
What each block looks like at each intensity
The block sizes scale roughly linearly with the new-card cap. The figures below are worked examples interpolated from the Anki anchor (20/day → ~200 reviews/day) and a 5 to 8 second per-card review time inferred from industry summaries.1212 They are arithmetic estimates, not measurements.
| Rate | Morning (new) | Commute (review) | Evening (cleanup) | Daily total |
|---|---|---|---|---|
| 5/day | ~5 min | ~5 min | ~5 min | ~15 min |
| 10/day | ~10 min | ~10 min | ~15 min | ~35 min |
| 15/day | ~10 min | ~15 min | ~20 min | ~45 min |
| 20/day | ~15 min | ~20 min | ~25 min | ~60 min |
When you only have one block
The reviews-first fallback follows from SM-2 lapse mechanics. An unreviewed card retains its scheduled interval but is now overdue, which under default SM-2 grading drives a lapse on the next attempt (the longer the overdue lag, the more likely the lapse).7 Unintroduced new cards are still in tomorrow's pool with no scheduling cost. In a constrained block, reviews are time-pressured and new cards are not.1
Reviews always take priority. New cards can wait.
The fallback inverts when accuracy is at risk: an additional new-card freeze, not a reviews-first squeeze, is the corrective. The burnout section below covers that threshold.
Tool-supported routines
The article's spine is tool-agnostic. The four numbers and three blocks above apply to any SRS that schedules reviews on roughly SM-2 or FSRS lines. Specific tools differ in configurability, opinionated pacing, and per-block ergonomics.
Anki: full control, full responsibility
Anki exposes both deck options the routine depends on: "New cards/day" (the cap on introduction rate) and "Maximum reviews/day" (the cap on the steady-state queue).1 Either can be set per-deck. The recommended pattern is to budget both together rather than treating "Maximum reviews" as an overflow valve.
Anki ships FSRS as a selectable scheduler from version 23.10 onward, with default desired retention 0.90.161 FSRS scheduling produces approximately 20 to 30% fewer reviews at the same retention compared to SM-2.10
The full-control trade-off is that deck content, card-template design, image and audio sourcing, and FSRS parameter tuning are the learner's responsibility. The Anki manual documents the options but does not prescribe the pacing.1
WaniKani: opinionated pacing, fixed order
WaniKani fixes the new-card delivery. The apprentice-stage SRS timings are 4 h → 8 h → 1 d → 2 d before promotion to Guru 1 on Levels 3 and above, accelerated to 2 h → 4 h → 8 h → 1 d on Levels 1 and 2.6 The full SRS stage sequence is Apprentice 1 to 4, Guru 1 to 2, Master, Enlightened, Burned, with intervals 4 h, 8 h, 1 d, 2 d, 1 w, 2 w, 1 mo, 4 mo, then permanent.6
The learner does not set a new-cards-per-day cap directly. The rate is set indirectly by how aggressively new levels are unlocked. Each level introduces about 5 to 7 radicals, about 30 kanji, and over 100 vocabulary items. The per-day average over a 12 to 18 month completion run sits in the 15 to 25 items/day band.6
WaniKani's check-in cadence is implicit in its early intervals: Apprentice 1 → 2 fires after 4 hours, which assumes the learner returns to the app multiple times per day. The three-block schedule is therefore functionally a fit for the WaniKani interval geometry, not just for self-built decks.6
Amenokori: the mobile-friendly fit for the three-block routine
For the morning, commute, and evening rhythm this article advocates, J-Compass recommends Amenokori. It is the SRS app whose shape lines up most directly with that three-block routine, and three features carry the fit.
First, mobile-first design. Amenokori is built around iOS and Android sessions with cloud sync, which fits naturally into the commute block: each session is short, interruption-tolerant, and resumable from another device in the evening block.185 A reviews-only block of 5 to 15 minutes on the phone is the use case the app was shaped around.
Second, FSRS scheduling as the only scheduler.185 FSRS produces the flatter standing-wave amplitude documented earlier in the article. That suits a learner whose evening block has hard time bounds and who cannot extend it if the queue overshoots.10
Third, daily caps that match the article's math. The free tier fixes "20 new cards and 150 reviews per day."5 The 20 new/day cap matches the Anki-manual canonical illustration of the high-end self-study rate.1 The 150 reviews/day cap sits inside the 150 to 200/day steady-state band the SM-2 ratio predicts at the 15 to 20 new/day intensities.
For the article's worked schedules from 5/day through 15/day, the free tier is sufficient without bumping into either cap. At 20 new/day the free tier hits the review cap exactly at the point where steady-state arrives.
Pre-built JLPT-level decks remove the 30+ minutes per day of deck-building, card-formatting, and FSRS tuning that Anki users absorb before the first review. That setup work is the largest fixed cost in starting an Anki-based routine.185
For the SRS-tool feature comparison itself (algorithm differences, deck ecosystem depth, pricing tiers), the choose-your-resources article is the right destination. This section does not re-litigate Amenokori versus Anki feature matrices. It names the fit between Amenokori's daily caps and the article's three-block math.
How to size the new-card cap inside any tool
The decision rule follows from the standing-wave math: choose the new-card cap that produces a steady-state queue you can sustain on a typical weekday, not on the best weekday.12 Because the standing wave does not arrive until weeks 4 to 6, a cap that feels comfortable in week 1 is not yet evidence that the cap is sustainable.
The reverse signal is a queue rising two days in a row. This is the standard early-warning indicator. It precedes accuracy collapse by approximately one to two weeks under SM-2's lapse-and-reset behaviour.7
The corrective is to cut new cards by at least 50% until the queue clears, not to raise the maximum-reviews cap. Raising the cap accommodates more lapses, but lapses are the symptom. The cause is the new-card rate that produced them.71
The burnout failure mode and how to detect it early
Kanji burnout is not just a mood. It is a measurable dynamic with three signals, a non-linear escalation pattern, and a structured recovery. Catching it in week 3 calls for a different routine than catching it in week 8.
The three early signals
Signal 1 is a queue that does not clear by bedtime. Under SM-2, an uncleared card overdue by at least one day is more likely to lapse on the next attempt. An overdue queue therefore feeds tomorrow's larger lapse pool, which feeds the next queue.7
Signal 2 is accuracy on first-pass reviews dropping below ~85%. The Anki manual quantifies how retention drives interval length and therefore queue size. Below 85%, the SM-2 ease distribution drifts downward (EF decreases by ~0.2 per low grade, with a floor of 1.3), and the queue per introduced card increases beyond the steady-state ratio.71
Signal 3 is sessions becoming chore-like. Community reports from the WaniKani burnout thread document this signature directly. The original poster (Level 27, 9.5 months of study) reports accuracy declining from 90%+ on average to around 85% at best and items resetting to apprentice-to-guru status repeatedly.19
Any one signal warrants a freeze on new cards. The combination of all three is the burnout signature documented across years of community threads.1920
The "queue debt" spiral
Queue debt is the snowball that follows from skipping reviews. Under SM-2, an unreviewed card keeps its scheduled interval and is then more likely to lapse on the next pass. That lapse resets the interval to 1 day with a lower easiness factor.7
Community evidence documents the same pattern. The "I skipped a few days and now i have almost 600 reviews" thread reports one skipped day producing about 150 reviews on return, with ~600-review backlogs requiring approximately a week of cumulative skipping.20 The dynamic is non-linear because the SM-2 schedule keeps producing reviews from older cards while the learner is also skipping the queue from newer ones.
The community remediation pattern is to batch the backlog in small chunks (for example, "20 reviews at a time"), sort lower-SRS-level items first (because their next interval is shortest and they will return again soonest if lapsed), and pre-emptively use WaniKani's vacation mode or Anki's deck-options reschedule before the next predictable absence.20
The intervention: a 7 to 14 day new-card freeze
The mechanic is to pause new-card introduction (set "New cards/day" to 0 in Anki, or do no new lessons in WaniKani) while existing reviews continue. The standing-wave amplitude is held constant while the existing queue drains through its scheduled intervals.716 After 7 to 14 days, the day-1 / day-6 / day-14 interval layers have all fired once for the most-recent cards introduced before the freeze. The queue stabilizes at the SM-2 steady state for the in-flight cohort.
On resume, restarting at half the previous new-card rate prevents an immediate return to the standing wave that produced the collapse. The new wave has half the amplitude, and the lapsed cards from the prior wave have been triaged.71
Community remediation patterns from the WaniKani burnout thread match this protocol: "reduce new lessons to 5 per day while maintaining reviews," "shift focus to grammar and reading practice rather than pursuing rapid level progression."19
When to take a full break instead
The threshold is accuracy below ~70% or 4+ consecutive missed days. At under 70% retention, FSRS workload analysis treats the schedule as inefficient: relearning costs dominate, and the optimal-retention page recommends optimizing the workload/knowledge ratio, which collapses below 80% retention.17 At 4+ consecutive missed days, queue debt under SM-2's lapse-and-reset behaviour usually exceeds what a freeze can drain without resetting most of the deck anyway.7
Community evidence supports the multi-week reset: the WaniKani burnout thread reports a level-60 finisher who took almost a year before feeling like starting up again, with retention "not that great, even on 'burned' items" on return.19 A clean break of 1 to 4 weeks, rather than an indefinite hiatus, plus a restart at the 5/day baseline is the structured version of this user-reported recovery.
Why "powering through" makes it worse
Forced reviews under fatigue produce lower-quality grades (more lapses), which under SM-2 increase the lapse pool (interval resets to 1 day, EF decreases), which inflates tomorrow's queue, which extends fatigue.7
The desirable-difficulties framework qualifies this directly: difficulty is desirable when it can be overcome through effort, not when it exceeds the learner's manageable range.9 Fatigue-zone reviews exceed that range and convert spaced practice into massed-with-relearning practice, which the distributed-practice literature documents as inferior to spaced practice at matched total study time.34
How the routine evolves over the year
A routine should not stay frozen at day-one settings. The standing-wave geometry, the habit-formation curve, and the cumulative-coverage curve each impose a phase transition at a known point.
Weeks 1 to 4: encoding phase
Standing-wave amplitude has not yet stabilized. The day-1, day-6, and day-14 layers are firing, but the day-35 layer has not.7 The daily queue at week 1 of a 20/day routine is roughly the new-card rate plus yesterday's day-1 reviews (~40). At week 3, it is roughly the new-card rate plus the accumulated day-1, day-6, and day-14 layers (~100). The full ~200 steady state arrives at weeks 4 to 6.1
The rule for this window is to hold the new-card cap fixed. The visible queue undersamples the true cost.
Weeks 4 to 12: steady-state phase
The standing wave has reached steady state. The queue size now reflects the actual sustainable load.712
Downward adjustments (reducing the new-card cap) happen in this window. Upward adjustments rarely survive the next four-week cycle.
Months 3 to 9: the long middle
The Lally habit-formation curve plateaus at a median 66 days (range 18 to 254) for habit automaticity.14 Months 3 to 9 sit inside this plateau for most learners.
Cepeda et al. 2008's temporal-ridgeline result is the key principle here. For a test one year away (such as a JLPT exam), the optimum inter-study gap is approximately 5% of the test delay, which corresponds to roughly weekly to bi-weekly intervals at this phase.3 SM-2's default schedule lands inside this range for cards in their second to fourth review.7
The temptation in this phase is to add Anki vocab decks, immersion, and grammar drill simultaneously. That is the secondary burnout source: each new deck contributes its own standing wave.
Endgame: tapering new cards as you approach your target
The cumulative-coverage curve flattens past the top-2,000 rank. Under the Bunkachō 2007 newspaper survey, the jōyō set covers ~99% of kanji tokens. The additional ~466 kanji needed to reach 99.9% buy ~0.9 percentage points.15 In the Asahi 1993 corpus (Chikamatsu et al.), the top 1,600 reaches 99% of kanji tokens. The rank 100 to 1,000 band carries the bulk of the coverage gain.21
Cards 1,800 to 2,136 of the jōyō list therefore add only marginal cumulative coverage. A pace cut (from 10/day to 5/day) in this band reallocates the review-time budget to vocabulary, reading, and immersion without measurable coverage loss.2115
Good to know
The wrong number to optimize is "kanji per day"
Optimize for the steady-state review load that can be sustained at 90% accuracy. Back-derive the new-card rate from that ceiling. The Anki ratio (20 new produces ~200 reviews/day at steady state) makes this conversion simple: choose the review minutes you can spend on a normal weekday, divide by the per-card review time (5 to 8 seconds), and use the resulting cap as the rate to set.12
Vocab decks and kanji decks share a review queue in your head
SM-2 and FSRS schedule each deck independently, but the learner reviews them in the same daily session and experiences the queues as additive. A 10-kanji/day routine that adds a 10-vocab/day deck roughly doubles the steady-state review load to ~200 cards/day under the Anki ratio.1
The failure mode is the learner who counts only the kanji deck and is surprised by 250 daily reviews. The cap is 20 across all decks at the upper end of the rate ladder, not 20 per deck.
Treating "maximum reviews/day" as the planning quantity
The Anki manual exposes two caps: "New cards/day" and "Maximum reviews/day."1 The temptation is to set "Maximum reviews/day" to whatever feels manageable and let the new-card cap float.
That reverses the real cost structure. "Maximum reviews/day" caps the visible queue but does not cap the standing-wave amplitude. Cards over the cap are deferred to tomorrow, and the standing wave compounds across the deferral. The new-card cap is the one that controls long-run load.1
"Powering through" framing in study communities
Community language ("just grind," "push through the queue") encourages massed review under fatigue. The Bjork and Bjork framework is explicit that difficulty must be manageable to be desirable. This framing risks pushing learners past the manageable range, where the spacing-effect literature predicts retention loss rather than retention gain.349
"Reviews are the rent, new cards are the mortgage"
New cards are the one-time encoding investment. Reviews are the recurring maintenance cost that comes due on a schedule the learner does not set. Treating the new-card cap as the routine's controllable lever (because reviews follow deterministically from it under SM-2 or FSRS) follows from the Anki manual's stated workload relationship.110
"FSRS" stands for Free Spaced Repetition Scheduler
The "Free" in FSRS is part of the algorithm's name, not a price claim. FSRS is the open-source successor to SuperMemo's SM-2 family, trained on memory-grade data with machine-learning techniques. The open-spaced-repetition benchmark documents FSRS outperforming SM-2 on recall prediction across 99.5% of 9,999 tested Anki collections.1610
The week-4 surprise
A learner who has held a new-card rate for three weeks without queue trouble has not yet seen the day-35 interval layer arrive. The standing-wave geometry predicts a near-doubling of the queue across weeks 4 to 6, even if nothing about the routine changes.71
Anticipating this transition prevents the false-confidence trap. The week-3 queue is not the steady state.
Stroke order practice does not have to be in the daily routine
Recognition reviews are screen tasks. Handwriting practice is a different motor skill on a different schedule. Bolting kanji-writing drills onto every daily session inflates per-card review time without changing the SRS standing wave.
A separate weekly handwriting block (or a separate week, for learners who care about handwriting at all) is a cleaner allocation than a per-session add-on. Recognition and handwriting have different intervals and different fatigue limits.
Weekend "catch up" is a tax, not a strategy
The failure pattern is weekday backlog pushed to Saturday. The math is unfavourable: weekend recall is worse because the gap inflates the average inter-review interval beyond the SM-2 optimum for the card's current stage. That pushes more cards into the lapse-and-reset pool.73
The fix is to lower the weekday new-card rate, not to schedule a weekend dump. A 5/day weekday rate that survives Monday to Friday produces a lower queue than a 10/day weekday rate that requires Saturday cleanup.
A routine that works at N5 will not work at N3
The readings layer compounds review-load math. A 10/day rate that was fine at N5 may produce roughly 1.5× the review queue at N3 because each new kanji surfaces two to four readings instead of one: on'yomi (Sino-Japanese readings), kun'yomi (native Japanese readings), and the occasional jukujikun or nanori. Each reading is a separate retrieval path the deck has to maintain.
The rate ladder above is therefore conditional on what counts as one item. A kanji-with-all-readings deck at N3 should be counted as roughly 1.5 to 2 items for every one item in a kanji-with-meaning deck at N5.
The mnemonic-authoring tax is not free
A Heisig-style self-authored mnemonic adds 30 to 60 seconds per new card on top of review time. At 10 new/day, that is another 5 to 10 minutes per day, budgeted separately from the new-card encoding minutes and the review minutes.
A curated mnemonic bank carries the load instead, at the cost of less personal salience. The trade-off is real and worth budgeting explicitly rather than absorbing as hidden overhead.
Sleep gates retention; review timing is secondary
A 10/day routine on 5 hours of sleep retains worse than a 5/day routine on 8 hours of sleep. If you must choose between the daily review block and an extra hour of sleep, cut new cards before cutting sleep. The exact mechanism is outside the scope of this article, but the direction of effect is consistent across the cognitive-psychology literature on sleep-dependent memory consolidation.
See also
- How to Learn Kanji: A Strategic Overview of Heisig, WaniKani, and Kanji-in-Context
- Your First Daily Japanese Study Routine: A Beginner's Template
- Beyond Anki: SRS Tools and Approaches Compared
- How to Get Back into Japanese After a Break: Recovering from Study Burnout
- How Long Does It Take to Learn Japanese? Setting Realistic Goals and the One-Year Trap
- Should You Learn Kanji in Frequency Order, School Order, or Pedagogical Order?