Study 30 words. Read your first Japanese story this week.

Contents

Every Japanese learner knows the deal you're offered: grind flashcards for six months, then you get to read something. The vocabulary app and the "real Japanese" are different worlds, and the bridge between them is a long, lonely walk on willpower.

I think that deal is backwards, so I built nerdreps around a different one: study your first thirty words, and I'll hand you a story you can actually read. Not a story with a dictionary bolted on, but a story where you already know the words, because it was written from your side of the bridge.

If you read the introduction post, this is the maybe I left hanging at the end of it. I called the problem chicken-and-egg there: input-based learning works, but you can't learn from input you don't understand. And I wondered whether the on-ramp had to be entirely vocab SRS, or whether it could be combined with reading. It can. Here's how.

The mechanic

nerdreps is a spaced-repetition app at heart: the first 1,000 Japanese words by frequency, scheduled with FSRS so every review lands right before you'd forget. That part is table stakes: Anki with a good core deck will do it, and Wanikani does it with more polish than I'll ever have. If those are working for you, keep going. nerdreps earns its keep on the other side of the flashcards.

The part I care about is the Stories tab. Every story ships with its vocabulary mapped word-for-word to the deck, and the app knows exactly which words you know, from your actual review history rather than a self-assessment. A story appears on your shelf when you know at least 95% of its words, with a hard rule on top: never more than one unknown word in a sentence.

That unknown word isn't a leak in the system. It's the point. Each story deliberately carries one word from the levels you haven't studied yet, so reading previews what study is about to teach, and study pays off what reading made you curious about. The two halves feed each other: your study dashboard tells you 「先生と学生」 is three words away, a new flashcard tells you this word appears in tonight's story, and tapping an unfamiliar word while reading queues it as tomorrow's first new card.

If you've read Stephen Krashen, you'll recognize all of this (comprehensible input, i+1, low-anxiety reading) made mechanical. There are no quizzes after the stories. No comprehension checks, no timers, no scores. You read, you tap a word if you need it, you read it again if you liked it. Rereading is studying; it just doesn't feel like it.

What thirty words buys you

Level 1-3 Japanese is all nouns (no verbs until level 5, no adjectives until 8), so the earliest stories are picture-book telegrams, and they lean into it:

朝です。雨です。 犬は家にいます。 猫も家にいます。 犬と猫は友達です。

Morning. Rain. A dog, a cat, a friendship. It isn't literature. But if you started studying this week, you just read Japanese, every word, with no dictionary. Something in your brain files that away very differently than a completed flashcard session.

By level 6 the stories have verbs and a dog who waits at the station every day for a friend everyone says isn't coming back. By level 20 they have first-person narration. 僕 arrives at that level, and the story that uses it ends: 僕の世界は小さいです。でも、犬がいます。 問題ないです。 My world is small. But there's a dog. No problem.

There are 37 stories today, from three-word sentences to a Tokyo trip, most of them threaded with recurring characters so the vocabulary recycles naturally. A native speaker copy-edited the lot. And yes, the dog at the station gets his ending, but you'll have to study to band 22 to read it.

The nerd section

A few things under the hood I'd want to know about if this were someone else's project:

  • The scheduler is FSRS-6, verified against the reference implementation (ts-fsrs) across a few hundred test points to within floating-point rounding. Card state is an event-sourced fold, so your history re-derives deterministically on every device, and can someday drive per-user optimized parameters.
  • Reviews are typed, not multiple-choice, and every word has a forgetting-curve chart: FSRS's predicted recall over time, drawn from your actual review history. The nerdy metrics I wished for in the intro post exist now.
  • Every sentence is narrated, in three voices: 夜語トバリ via VOICEVOX for natural pitch accent, plus two Kokoro-82M voices generated locally. Read-along highlights the sentence being spoken; playback speed and furigana are toggles.
  • It's local-first. No signup to start; progress lives in your browser, syncs if you make an account, works offline as a PWA. The free tier (three levels and six stories) is free forever, no card, and the stories stay yours to reread.
  • Stories are validated, not vibed. A build-time validator enforces the coverage floors and the one-new-word rule against the deck before anything ships. If a story is on your shelf, the math says you can read it.

Where this is going

nerdreps is early: one deck (Japanese), 37 stories, and a roadmap that includes scene illustrations for the earliest bands, more narrators, and stories climbing toward the full thousand words. One paid upgrade unlocks all levels and every future Japanese content update; the price of a couple of textbooks, once, and the free tier is a real education on its own.

If the idea of an SRS that pays you in stories sounds like your kind of deal: nerdreps.com. Start with 朝です。雨です。 and see how it feels to read.

Feedback, bug reports, and opinions about the cat's personality: I want all of them.

for now you can contact me at [email protected]

--
Sprocket