{{first_name | みんな}}さん、こんにちは ☕

It's a Sunday morning, seventeen days before the test. You open your SRS app. The due count is three hundred and twelve. By card forty, the same five words you've been failing since March cycle through again — 落おち着つく, 仕方しかたない, 都合つごう, 結局けっきょく, 一応いちおう — and you fail four of them, again.

Forty-seven minutes in, you look at the clock and realize the section you're actually weakest on hasn't been touched.

Then the thought arrives. Maybe I should just drop the leeches. Then the counter-thought, instantly, louder: but what if 落ち着く shows up in the reading passage?

So you keep grinding. You will do the same thing tomorrow.

Here's the move nobody hands you. Every card in your pile belongs in exactly one of four buckets — Drop, Defer, Drill, Deploy — and only one bucket gets your real attention for the next seventeen days.

The card you suspend today is not the card that beats you on test day. The card that beats you is the fragile one you kept brittle by spreading your attention across six hundred cold items.

That is the whole reframe. We will spend the rest of this issue on the rule, the buckets, the three-second test that sorts them, and a twenty-minute pile-sort you can do this Sunday.

Two notes before we go. We've zoomed out on the macro plan in The 6-Week JLPT Final-Push Plan — this issue is the vocab-triage zoom-in. And the reflex work we did last week in The 即時応答 Rescue lives downstream of the same idea: at T-17 days, retrieval speed on the things you almost-know beats coverage of the things you don't.

Permission to drop is the missing teaching here. We want to say that out loud before the sort, because the rule is useless if you can't grant it to yourself.

Let's get the rule on the table. Then we'll walk a card through each bucket.

Drop. The card has failed many times (Anki's default tags a "leech" at 8 lapses — apply the equivalent in your SRS) AND the word sits outside your level's high-frequency block. Suspend it. The agonizing is the trap. These cards aren't crossing threshold in seventeen days; trying will cost you the cards that would have.

Defer. The card is new — you've never answered it correctly — and it's outside the high-frequency block. Move it to a "post-July" deck. Suspended for now. It comes back July 6 like nothing happened.

Drill. The card is fragile. You recognize it slowly. You can sometimes recall the meaning but not the reading, or the reading but not the meaning. This is the bucket that wins or loses your test. Sixty to seventy percent of your daily vocab time goes here.

Deploy. The card is reliable. Recall under three seconds. Multiple successful retrievals in a row. You'd bet money on it. Light spaced touches only — ten to fifteen minutes a day. The goal isn't to learn it. The goal is to keep it warm.

A receptive-vs-productive note before we go further, because this is where the rule earns the reframe.

You almost certainly recognize more Japanese than you can produce. The vocabulary literature (Webb 2008; Laufer 1998; with Schmitt 2014 and Nation 2013 explicitly cautioning against any universal multiplier) puts the surplus somewhere between roughly 1.1× and 2×, depending on proficiency and how strictly "production" is measured. The exact number matters less than the direction. The JLPT only tests the recognition side. A card you can match to a meaning when you see it on the page is a JLPT-ready card, even if you couldn't produce it from English in a million years. That is permission, not a corner-cut.

Now the three-second test. Apply it to every card during triage.

See the front. Count out loud: one Mississippi, two Mississippi, three Mississippi. (Or whatever three-second beat works in your head.)

Counting out loud matters. Silent counting is where the "I almost knew it" lie sneaks back in and contaminates the sort.

A note on the frequency bands, because we're going to use them in the next breath and we owe you the disclaimer first. There is no official JLPT vocabulary list. The Japan Foundation has never published one. So when we say "high-frequency block," we mean a working heuristic that almost every prep tool already uses: rank the card against a published frequency list (jpdb.io is the cleanest, or Tanos's lists, or the Core 6k/10k decks). For test-takers aiming at the intermediate band, drop candidates sitting outside roughly the top three thousand most-frequent words. For upper-intermediate, the working line is around the top six thousand. For advanced, around ten thousand. These are planning targets, not boundaries the JLPT enforces.

Now the four cards. Live examples, one per bucket.

The default SRS algorithm optimizes long-term retention given infinite time. You don't have infinite time. You have seventeen days. The optimization function changed three weeks ago and nobody told the algorithm.

So the time allocation has to change too. Roughly:

What we just bought you, versus grinding a ninety-minute leech queue, is somewhere between thirty and forty-five minutes a day. The swing depends on how big your queue actually is — readers with a hundred due cards will land at the lower end, readers with six hundred at the higher one. Over seventeen days, that's roughly four to seven hours of attention you can move to mock-test pacing, the listening drill from last week, or the reading-speed work from the issue before.

That is a score move. It is also, structurally, what consistency-over-intensity actually looks like in the last stretch. Twenty minutes of triage today plus thirty minutes of focused drill per day for seventeen days beats one panicked Saturday binge through six hundred cold cards. The literature on retrieval practice (Roediger & Karpicke 2006; the Adesope et al. 2017 meta-analysis puts the effect at around g ≈ 0.50) is clear that retrieval reps compound on cards you almost know. They don't compound on cold cards. Almost-knowing is the precondition. The Drill bucket is where almost-knowing lives.

The 20-Minute Pile Sort — do this once, this Sunday June 22 at the latest.

Open your primary vocab deck (Anki, Bunpro, Renshuu, jpdb — whichever has your largest backlog). Set a 20-minute timer. For the first fifty due-or-new cards in your queue, apply the three-second count-out-loud test and drop each card into a bucket using the rule from the Deep Dive. At the timer, stop. Count your buckets.

(No literal answer key — we'll share the cohort distribution and a tactical follow-up in next week's brew. Try the sort first.)

You'll know you did it when: You can open your SRS app and every card you see is one you've explicitly decided is worth your attention today. No leech cycling. No mid-review agonizing. Every card is in a bucket, and the bucket has a behavior.

Last week's answers (The 即時応答 Rescue): No literal key — that one was a reflex drill. Three patterns to flag, based on the brief's pedagogy and the early replies. Pattern 1: 確認かくにん (confirmation, the 〜だったっけ form) is the most-missed speech-act category. Most learners parse it as a question and reach for an answer reply when the prompt wants a confirmation reply. Pattern 2: 申もうし出で (offer) and 勧誘かんゆう (invitation) collapse together around 〜ましょうか. The fix is asking "is the speaker doing it alone, or proposing we both do it?" Pattern 3: pragmatics errors outnumber parsing errors at the plateau levels. The fix isn't more grammar. It's ten more reps of the function-to-chunk match.

If you only sort twenty-five cards tomorrow morning and suspend the ten that have been failing since March, you've already done more vocab triage than most of the cohort will between now and July 5. That's the bar. One bucket, one decision, one less card cycling through your reviews tomorrow.

For: Plateau fighters who want a frequency reference to rank their cards against during the pile sort.

Use: jpdb.io — paste any Japanese word, get its frequency rank against a real corpus in two seconds. Use it during Sunday's triage as your high-frequency / low-frequency check.

Why it helps: The bucket rule depends on knowing where a card sits in frequency. Without that anchor, "drop the low-frequency leeches" stays abstract. With it, the call is mechanical. If you don't want a new tool, the Core 6k/10k Anki deck order works the same way — anything in the deck is high-frequency by definition.

これからも一緒いっしょに頑張がんばりましょうね〜 💪