{{first_name | みんな}}さん、こんにちは ☕

We mined 15 sentences from one anime episode. Felt great. Tomorrow-us hated past-us.

Six weeks later, 12 of those 15 were leeches. Eight fails each, zero sightings in the wild. The hard work was real. The output was noise.

The problem wasn't sentence mining. It was sentence mining without a filter.

Know someone with 3,000 cards and zero progress? Forward this — they need it more than us.

Most sentence mining advice is fine. The selection criteria is what's broken.

Here's the mistake almost every immersion guide tiptoes around: they teach you the shape of a good mining sentence, not whether a particular sentence is worth your time.

The immersion community's "one unknown word per sentence" rule is a useful simplification of Krashen's i+1 — but it's a format check, not a selection check. Every page of native content has thirty sentences that pass the format check. Mining all thirty is what wrecks your deck.

So here's the reframe we want you to carry into your next session.

One adds cards. The other fixes leaks.

Last week's edition is a perfect example. If ている confusion has been quietly tripping you up — you read 死しんでいる as "is dying," you hesitate before 結婚けっこんしている — that's the gap. Mining a sentence that nails the exact pattern you keep getting wrong is gap-mining at its purest. You noticed the leak. You found the fix. The card almost can't fail.

Mining 適当てきとう because it showed up in an anime and you didn't know it? Coverage. The card joins the pile. Six weeks later it's a leech.

The difference isn't the sentence. It's the decision in front of it.

We've been there. We've sat in front of Yomichan at 11 PM telling ourselves "well, I didn't know it, so it counts." It doesn't. Not knowing a word is the floor for mining, not the ceiling.

Both sentences below are valid i+1. Both pulled from the same chapter. Only one becomes a card.

The difference: A targets a pattern you'll use this week — られる + ていない for a state you keep meaning to express. B is a beautifully literary collocation you might meet again in three months. A is a gap. B is coverage. Mine A. Skip B (or, fine, screenshot it).

A becomes fluent. B becomes a leech with poetry.

If your deck has 3,000 cards, you don't have a vocabulary problem. You have a selection problem.

Quick gut-check: think of the last 10 cards you added. How many were "I keep failing this in real life" — and how many were "I didn't know it, so I grabbed it"?

That ratio is your filter, with or without one. Here's the deliberate version.

1. Gap. Have we noticed this word or pattern failing us in real use — reading, listening, or speaking? Not in the abstract. In the last two weeks. If we haven't bumped into a need for it outside this mining session, the answer is no.

2. Sentence quality. After the lookup, does the whole sentence make sense? Does the surrounding context — the scene, the audio, the panel — reinforce the meaning? If the sentence requires two lookups, or we can't reconstruct why it means what it means, the card will fail. Find a better example or skip the word entirely.

3. Frequency. Will we realistically encounter this in the next month of normal immersion? If it's a literary-only word and we mostly read manga, or a keigo form and we mostly watch anime, the ROI is near zero.

If any answer is no — skip. The card costs nothing to skip. Skipping is the move.

Mining is not studying. It's adding debt.

Before the math: which Future You wrote tomorrow's review queue? Past You did. Yesterday. With every card you waved through.

Cap mining at 5–10 new cards per session. Hard cap. Even if you find fifteen valid candidates.

This isn't willpower. It's math. We covered the multiplier in our SRS Triage edition: at steady state, daily reviews land around 10x your daily new cards.

Mining 10/day = ~100 reviews tomorrow. Mining 20/day = ~200. The bottleneck is never your mining hour. It's tomorrow's review queue.

20 new cards today is you scheduling a 40-minute punishment for Future You.

Capping the input is half the discipline. The other half is knowing when to stop entirely.

That's the input side. Now the off-ramp.

Every "maybe" card becomes tomorrow's punishment.

If you can read 80% without pausing and you're still mining 10/day, you're not learning Japanese — you're maintaining a database.

When comprehension lands around 80% without pausing, mining's job is mostly done. Shift weight toward pure immersion. Let frequency carry the rest.

This is the part most guides skip — when to stop. There's no "graduation."

There's a slow handoff: 95% mining / 5% reading at the start, drifting to 5% mining / 95% reading at the end.

And one more trap before we move on.

Quick gut-check: which genre are you accidentally training for right now? If every card you've made for the last three months came from one anime, you know twelve words for sword techniques and zero words for "the shipment was delayed."

Genre lock-in feels like a plateau. It's actually a coverage hole.

Rotate sources. Manga and a podcast. Novels and a news article. Not because variety is virtuous — because monoculture decks build blind spots that masquerade as ceilings.

This week: a Mining Audit + a 5-Card Session. 15–20 minutes total.

(No answer key on this one — the answers are in your own deck. Reply and tell us your audit ratio if you want us to react in next week's brew.)

You'll know you did it when: You created exactly five cards (or fewer) and can name the gap each one fills. You also identified at least one sentence you would have mined a month ago and chose to skip today.

That audit ratio — cards kept versus candidates — is the single most useful number in your study notebook this month. Track it. Watch it shrink.

If you read this far, you already know more about what to mine than most learners with three thousand cards. The frameworks tell you how to make a card. The filter tells you whether the card should exist at all. That's the whole shift.

For: Anyone whose mining setup needs a tune-up alongside the filter.

Use: Read Refold's Sentence Mining roadmap for the full mechanics — synchronous vs. asynchronous, card formats, the 1T rule. For the tooling layer (Yomitan + mpv + subs2srs), Tatsumoto's immersion toolset is the canonical reference.

Why it helps: We gave you the filter. Refold gives you the setup. The combination — what to mine plus how to mine — is the full picture.

This week, run the 3-Question Filter on your next mining session. Before every card: Gap? Good sentence? Frequency? Reply and tell us how many cards survived the filter — out of how many candidates. We read every reply.

If you're the kind of learner who actually finishes decks, send this to the friend drowning in leeches.

P.S. — Last week's Verb Sort Challenge: 走はしる, 読よむ, 泳およぐ, 書かく, 歌うたう are process verbs — ている means the action is happening right now. 結婚けっこんする, 落おちる, 届とどく, 壊こわれる, 始はじまる are point verbs — ている means the change already happened and the state remains. The pattern: if the action has duration, ている is progressive. If it's instantaneous, ている is resultative. If you got the process verbs right but stumbled on point verbs, you're in the most common bucket — and exactly the one the framework is built to fix.