I Am Learning Japanese. Here Is the Map (Month 0)

千里の道も一歩から Even a journey of a thousand ri begins with a single step.

Tomorrow, September 1st, I start studying Japanese. The deadline I gave myself is August 31st, 2029. Three years, thirty to sixty minutes a day, starting from absolute zero. And starting today I’m going to publish a monthly update here, every month, for thirty six months, so the whole thing is documented in public and I can’t pretend in month nineteen that I never said any of this.

This is Month 0, so there’s nothing to report. That’s the point of it. This post is the plan before contact with reality, and everything after it gets measured against what’s written here.

So why would a man living in Italy want to learn Japanese with no intention of ever living in Japan?

Japan Broke Something in Me

I went to Japan. And like roughly one hundred percent of the people who go to Japan, I came back different.

I’m not going to write the post you’ve already read fifty times. You know the one. The trains are on time, the toilets have opinions, nobody litters, the convenience store sandwich is inexplicably better than a restaurant meal in most of Europe. All of it is true and all of it has been documented to death by people with better cameras than mine.

What got me was something else, and it took me months to work out what it was: for two weeks I was functionally illiterate for the first time in my adult life.

I’ve spent my whole life in countries where, even when I didn’t speak the language, I could decode it. A sign in French, a warning label in Dutch. I don’t speak either, but the letters are the same letters, and with enough squinting and Latin roots I get maybe forty percent of the way there for free. Written language, for me, has always leaked meaning even when nobody invited me in.

In Japan nothing leaks. A sign is a wall, a menu is a wall, and the little handwritten note taped to the door of the tiny place you wanted to eat at, the one that would tell you whether they’re closed for the holiday or closed forever, is also a wall. You stand there like a golden retriever looking at a spreadsheet.

And here’s what has kept bothering me since: Japan was extremely kind to me while I was illiterate. The country is so well engineered for visitors that you can have a wonderful time understanding absolutely nothing. English signage in the stations, picture menus, translation apps, staff who will patiently mime an entire transaction at you. The infrastructure of tourism is good enough that being illiterate never becomes uncomfortable enough to register as a problem, so it never gets fixed. It just puts a ceiling on what you’re allowed to have, and you don’t notice the ceiling is there.

The Japan I Did Not Get to See

There were moments, though. There were always moments. The one that stuck: standing in a bookshop surrounded by thousands of things I couldn’t read, in a country whose comics I’ve been consuming my entire life. I bought a manga anyway. It’s been sitting on a shelf at home ever since, purely decorative, which tells you everything about the problem.

Tourism hands you the country as an exhibit. You get the temples, the food, the neon, the postcard version of Tokyo and Osaka and Kyoto, and it’s genuinely wonderful. But you’re looking at all of it through armored glass. Everything reaches you pre-translated and pre-approved for foreigners, and you’re never in the room, you’re only ever looking into the room.

What I want isn’t another visit. I want to go back and be in the room. To read the handwritten sign and find out they’re closed on Tuesdays. To end up in a town that’s on nobody’s itinerary, where there’s no English menu because there has never been a reason to print one, and order lunch, and talk about the weather with somebody who has no particular interest in me being there. To walk into that bookshop and leave with something I chose because of what was written in it.

So I want to rediscover Tokyo, Osaka and Kyoto with the glass removed. And then I want the rest of the country, because the parts of Japan that tourists never reach aren’t hidden. They’re just written in Japanese.

Yes, Fine, There Is Also Anime

I like anime. I like manga. I’ve played an enormous number of Japanese games, and anyone who has read this blog knows my Steam library is a war crime. A large share of the media I’ve loved in my life was made in Japan and reached me through a translation layer of extremely variable quality.

So yes, I want to play a JRPG in Japanese, and I want to read a manga volume the week it comes out instead of a day later with some unofficial translation, if at all. There’s also a part of me that’s annoyed on principle whenever the English localization of something I care about was clearly done by somebody having a rough week.

But if that were the whole motivation I wouldn’t be doing this. Media is a fine reason to learn two hundred words and a terrible reason to sign up for three years and roughly a thousand hours, because the moment the plateau arrives in month five, “I would quite like to read manga” evaporates on contact. Wanting to consume something has never once made me do anything difficult for three years.

Wanting to talk to people might. That’s the bet, anyway. Media is the reward at the end, not the engine.

There’s also a smaller and much pettier reason, which is that after Italian I got a bit cocky about languages, and Japanese appears to be the universe’s way of correcting that. The arithmetic section below contains my punishment, in a table.

First, a Detour: I Built the Thing That Built the Plan

My first instinct, being who I am, was to ask an AI for a study plan. Everybody does this now. And what comes back is lovely. Beautifully formatted, week by week, twelve recommended resources, encouraging milestones, the lot. It reads like a plan written by somebody competent.

It’s also almost entirely useless, for five reasons that took me a while to name.

• The week by week schedule assumes you’ll never get ill, never travel and never have a bad month. You’ll break it in week three and conclude that the failure is yours.
• It only ever adds. Twelve resources, all recommended, none cut. The hardest part of learning something alone is deciding what not to study, and that’s precisely the part it won’t help you with.
• The milestones can’t be measured. “By month three you will be comfortable with the basics.” Comfortable how? Measured against what? By whom?
• Feasibility is asserted rather than calculated. “Three years is plenty!” Is it? Show me the arithmetic.
• And it’s generic. The same plan for a monolingual beginner and for someone who already speaks four languages, even though those two people are going to fail in completely different ways.

So you feel organized, you buy four books, and you quit in month five never having found out whether the plan was wrong or you were.

I decided to fix the structural causes rather than complain about the tone. The result is a skill for Claude Code, which is a folder of instructions the model reads and follows, that I wrote and called Chizu.

地図, chizu, means “map”, and the whole idea is in the name: deliver the terrain before anyone takes a step. It doesn’t teach you anything. It designs the path and hands it back.

It runs six phases, one per response, each gated on the last:

Four things in there that a normal plan doesn’t do.

It maps interference, not just gaps. Most plans list what you lack. The more useful half is what you already know that’s going to actively mislead you, which for me turned out to be the whole story, and I’ll come back to it in the grammar section.

It also computes the budget instead of asserting it:

Available = minutes/day × days to deadline × adherence factor (0.85 to 0.90)
Required  = Σ (block hour estimates)

If those two numbers don’t meet, you find out before you start rather than at month thirty.

Every gate names its instrument. “Comfortable with the basics” is not a gate. “190 out of 200 on a mixed recognition quiz, 50 items in under 60 seconds” is a gate. If the instrument doesn’t exist yet, the plan tells you when you have to build it, normally before day one of that block, because a criterion nobody can measure decays into self assessment by vibes within about a month.

And it plans for you quitting. Months three to eight are the structural dropout window for more or less any multi month project, the stretch where the novelty has gone, the plateau has arrived and nothing visible happens from one week to the next. Chizu names that window up front and writes the relapse protocol before it’s needed and emotional. The rule I expect to need most: never repay missed sessions, because debt repayment is what causes the second and much larger relapse. What you’re protecting is the return, not the streak.

It’s here:

git clone https://gitlab.com/doismiu/chizu.git ~/.claude/skills/chizu

Then you say “I want to learn Japanese” and it refuses to plan anything until it has interrogated you properly. Two of its Phase 0 questions are ones I’d never seen a planning tool ask: what your adherence actually looks like on multi month projects, and what your dead time looks like. The first changed the design of my plan more than any other answer I gave. The second found twenty to thirty minutes a day of listening that had never appeared in my stated time budget.

You can also paste the body of SKILL.md into any decent chat model as a system prompt. The frontmatter is only there for skill autodiscovery.

A tool about honest measurement shouldn’t oversell itself, so the README carries a table separating what rests on well replicated findings (retrieval practice, spacing, interleaving, desirable difficulties) from what is structure imposed for practical reasons (the phase framework itself, the fixed interval ladders). The framework isn’t science. What it does is make the well supported techniques hard to skip, and make the plan’s own weak points visible instead of hidden.

The Japanese plan below is the first real thing it ever produced. Everything from here is its output, and it now lives in a git repository called senri, where every block I close becomes a commit.

The Learning, and the Grammar

If you’ve never looked into learning Japanese, none of the names below will mean anything, so here’s the short version before the plan uses them as though they’re obvious.

Genki is a beginner textbook in two volumes, published by The Japan Times, and it’s what most university courses in the West put in front of a first year student. It isn’t clever and it isn’t fashionable. It’s the boring correct answer, it comes with a workbook and a free audio app, and roughly a decade of internet arguments have failed to dislodge it. Quartet is the same idea one level up, for the intermediate years after Genki runs out.

So the whole plan collapses into one sentence:

GENKI is the course. Everything else is instrumentation.

Anki, audio, the recorder, the notebook and the AI exist to serve Genki. The day an instrument competes with Genki for the time in the core session, the instrument loses. I’m not following a plan and a book. I’m following the book, and the plan is just the map I open when I get lost.

The architecture, in one line:

GENKI I → GENKI II → QUARTET I → QUARTET II → native material
   ↑ teaches something
Anki         → stops me forgetting the vocabulary that matters
audio/speech → forces real time processing
Satori       → bridge to actual Japanese
native media → eventually replaces learner material

Which, in plain words:

And the day, entire:

1. Core. The work of the current block. 30 minutes guaranteed, 60 when possible.
2. Exposure. 20 to 30 minutes of audio in dead time, outside the core.

That’s it. There’s no third thing. If you’re wondering where the elaborate color coded timetable went, there isn’t one, and that’s deliberate.

The Nine Blocks

Progression is locked by mastery rather than by calendar. If Block 4 takes seven months instead of five, it takes seven months, and the criteria never get loosened to fit a date. When the calendar doesn’t fit, the calendar is what gives way.

Block 0: Sound Before Anything Else

I don’t start with hiragana. I start with listening, for seven to ten days, before a single character.

This felt wrong to me when I first read it, and I’ve come round to thinking it’s the best decision in the plan. If you attach vocabulary before your ear is calibrated, you memorize hundreds of words with the wrong pronunciation baked in, and then you spend years unpicking it.

Japanese isn’t rhythmed by stress the way Italian and Spanish are. It’s rhythmed by moras, which are beats. きて is き・て, two beats. きって is き・っ・て, three beats. Different words. And vowel length is contrastive, which is a thing none of my four languages does. Not one of them.

The instrument I built for this is a spreadsheet of thirty minimal pairs, which are two real words separated by exactly one sound, so that failing to hear the difference is failing to hear the word:

pair,contrast,priority,a_kana,a_romaji,a_moras,a_meaning,b_kana,b_romaji,b_moras,b_meaning
1,long vowel,1,おじさん,ojisan,4,uncle,おじいさん,ojiisan,5,grandfather
2,long vowel,1,おばさん,obasan,4,aunt,おばあさん,obaasan,5,grandmother
3,long vowel,1,ゆき,yuki,2,snow,ゆうき,yuuki,3,courage
9,long vowel,1,きて,kite,2,come,きいて,kiite,3,listen
11,long vowel,1,おい,oi,2,nephew,おおい,ooi,3,many
13,long vowel,1,ビル,biru,2,building,ビール,biiru,3,beer
15,geminate っ,2,きて,kite,2,come,きって,kitte,3,postage stamp
18,geminate っ,2,おと,oto,2,sound,おっと,otto,3,husband

Note pair 1. Get the vowel length wrong and you’ve changed which relative you’re talking about. Note pair 13, where I order a building instead of a beer, which is the kind of mistake that would end me.

The loop is fixed and the order isn’t negotiable:

hear the contrast → decide what you heard → reveal → hear again → imitate → record → compare

Never read the answer before hearing it. Reading first tests reading rather than listening, and it permanently ruins that piece of audio as listening training. This is the rule I expect to break first.

The gate to leave Block 0 is four boxes, all of which need ticking:

• Perception: ≥90% on controlled long/short and single/geminate contrasts
• Moras: correct count on ≥18 of 20 words
• Imitation: 15 to 20 short native sentences reproduced with the timing preserved
• Production: /r/, う, ふ, し, ち, つ distinguishable from the equivalents in my languages

A native accent is explicitly not required. This is calibration, not perfectionism.

Block 1: Kana, and Killing Romaji

More vocabulary first, because Japanese runs three writing systems at once and this section is unreadable without them:

So: two to three weeks. Hiragana, then katakana, then the contracted combinations, then typing on a Japanese IME, the input system that turns what you type on a normal keyboard into kana and then offers you the right kanji.

One rule dominates: romaji is permitted only at first contact with a character, and never again. After that it’s an answer key rather than reading material, and by the end of the block, reading Japanese in romaji should be less comfortable than reading kana.

With four languages already living in the Latin alphabet, romaji doesn’t sit lightly on top of Japanese. It sticks. It becomes the thing I actually read, with the kana reduced to decoration, and at that point I’ve learned a transliteration system instead of a language.

The gate is measured rather than felt: 190 out of 200 on a mixed kana test, at roughly one second per character, without visual analysis. And the plan is explicit about the failure mode I would otherwise talk myself past:

If recognition is correct but slow, Block 1 is not finished.

Grammar: Where My Four Languages Betray Me

This is the part of the plan I find genuinely interesting, because it inverts the usual thing.

Every study plan tells you what you lack. Mine spends more effort on what I already have that’s going to mislead me. I’m not arriving at Japanese empty. I’m arriving with four Romance and Germanic languages already installed, and they’ve given me a set of expectations that Japanese is going to violate one at a time.

What helps:

• The five vowel system of Italian and Spanish gives me clean Japanese vowels with none of the vowel reduction English speakers fight for years.
• The tapped /r/ of Portuguese and Spanish gets me most of the way to らりるれろ.
• Italian already taught me that consonant length changes the word, so っ is a calibration job rather than a construction job. Fato and fatto. I’ve done this before.
• English gives me access to nearly all the learning material and to thousands of loanwords.

What actively harms:

• I expect an explicit subject. Japanese omits the subject, the object, and anything else the context already supplies. My instinct will be to hunt for the missing piece, and there is no missing piece.
• I expect word order to carry the grammar. In my languages, position tells you who did what. In Japanese that job belongs to the particles, and the verb turns up at the end. This is the big one.
• I’ll overuse pronouns. 私 and あなた everywhere, because Romance languages trained me that people need naming. In Japanese, using あなた too much ranges from odd to actively rude.
• I have a stress accent instinct. I’ll want to lean on a syllable. Japanese doesn’t work that way, and what distinguishes the words is length, which my ear currently doesn’t hear at all.

Underneath all four sits the trap that will cost me most: treating Japanese sentences as encrypted Romance sentences, hunting for where the subject is and which word is the article. Japanese constructions have to be learned as Japanese constructions rather than decoded into Spanish and back.

The grammar itself starts in Block 2, which is Genki I lessons 1 to 6 over about two months, and covers topic versus grammatical role, the particles は, が, を, に, で, と and も, verb final processing, copular sentences, existence, basic verbs, how adjectives behave, past and non past, affirmative and negative, the polite and plain registers at a basic level, and the て-form, which is the verb ending that a large share of later Japanese grammar hangs off, here used for requests and permission. Around four to five hundred high frequency words go with it.

Speech starts here too, in Block 2, not later. That’s a hard rule: delay the paid tutor, never delay speaking.

The loop for every Genki lesson, always in this order:

The gate out of Block 2 isn’t a quiz. It’s two to three unscripted minutes about my own day, recorded. Pauses allowed, script not allowed.

Kanji, and the Trap Everybody Falls Into

The plan is aggressive here because this is where most self learners lose a year.

Kanji arrives through vocabulary, never as isolated characters. I will never sit down and study twenty kanji today.

What “knowing a kanji” means here is not this:

木 = tree, on reading モク/ボク, kun reading き/こ, reproduce from memory, memorize every compound.

It’s this:

日本 → にほん → Japan. 日本人 → にほんじん → a Japanese person. 日本語 → にほんご → the Japanese language.

Through those words, 日 and 本 become familiar characters carrying readings that actually occur. The count then builds itself:

0 → ~60 → 145 → 317 → 644 → 974 → ~1400 → ~1800+

The acceleration is deliberate. The five hundredth kanji is far easier than the twentieth, because by then vocabulary, grammar, component knowledge and reading experience are all holding it up, which is exactly why you don’t front load them. “First I’ll learn the kanji, then I’ll learn Japanese” burns years. Japanese children already speak Japanese when they learn to read, so they’re solving a completely different problem to mine.

WaniKani is out for the same reason. It would create a second kanji progression running out of sync with my actual curriculum, and when two progressions cover one function, each one permanently makes the other feel like failure.

The Arithmetic, Which Is Where It Gets Uncomfortable

I would never have produced this section on my own, because producing it meant finding out something I didn’t want to know.

Three years is 1,095 days. With an adherence factor of 87%, which works out at roughly 47 days a year lost to travel, illness, work and simply bad weeks:

And the cost of the plan, adding up the block estimates: 805 hours at the floor, 1,088 at the ceiling.

Verdict: viable, but tight. Not comfortable.

Which means the honest reading isn’t the comfortable “45 to 60 minutes a day”. It’s that 45 minutes is the floor to finish Block 7, and 60 minutes is what the plan actually requires. Thirty minutes stays as the minimum for bad days, but now with the explicit knowledge that a long run of thirty minute days moves the schedule for real rather than symbolically.

I asked for a study plan and got told to do the math, which wasn’t the emotional experience I’d signed up for, but there we are.

Then it went further and calibrated the outcome. The FSI, which is the US State Department’s language school and has been timing this sort of thing since the 1940s, puts Japanese in its hardest category, around 2,200 hours of intensive instruction for professional proficiency. So with roughly a thousand hours over three years:

That’s about 75 to 80% of what I said I wanted. And I’d much rather have that number written down now than discover it in month thirty and read a completely normal result as personal failure.

What I Am Actively Ignoring

Japanese offers nearly unlimited opportunities to study things that look productive. So, explicitly discarded:

The rule underneath all of it: don’t optimize for completeness, optimize for functional coverage.

And the biggest single risk to this whole plan isn’t my motivation. It’s Anki. At twenty new cards a day the review queue stabilizes at 150 to 200 cards a day within a few months, which on its own eats the entire thirty minute core, and then everything else dies. So: ten new cards a day, a hard ceiling by time rather than by queue, new cards drop to zero whenever the queue grows three days running, and any card I’ve failed four times gets deleted. Deleting a card is a normal and healthy operation rather than surrender.

The Next Ninety Days, and Then Every Month

The first thirteen weeks, from tomorrow to the 30th of November, cover Blocks 0, 1 and 2. Week 1 is sound. Weeks 2 to 6 are kana. Week 7 is where Genki lesson 1 and Anki both begin. By week 13 the target is an unscripted two to three minute monologue, recorded.

Measurement happens every four weeks, and what it records is behavior rather than feeling. Not “I feel like I’m progressing”, but seconds per kana, percentage on sound contrasts, seconds until I can form a spoken sentence, how long I can talk before I stall, lookups per ten minutes of reading. Numbers I can be wrong about in public.

That’s what this blog is for now. On the first weekend of every month I’ll publish the measurement, whatever it says. If the Block 1 gate fails in week 6 and everything slides into December, that goes in the post, because a gate holding is the plan working. The failure would be declaring kana finished in order to protect a date.

Two things I expect to write about, and I’d like them on the record now so nobody can accuse me of retrofitting them later. Weeks 9 and 12 are the heavy ones, Genki lesson 3 with the first verbs and lesson 6 with the て-form, and both routinely cost self learners more than the week they’re given. And months three to eight are the dropout window, roughly February to July next year. If these posts get thin around then, you’ll know exactly what you’re looking at.

The repository is called senri, after 千里, the thousand ri of the proverb. The current state section of the README gets updated at the close of every block, which turns the git history into the record without me having to keep one. And whether the whole thing works or falls apart, at least it’ll be instrumented when it does.

Anyway. See you in a month, when I’ll have spent thirty days listening very carefully to a man saying おじさん and おじいさん at me, and will be able to tell you, with a percentage, whether I can hear the difference between my uncle and my grandfather.

じゃあね!