passed.jp

English grammar is boring. You can be told "the be-verb agrees with the subject" a hundred times and still not remember it. But the single sentence "my husband was killed by a giant anteater" is something you can't forget once you've read it.

The "Impact" department we built on english.passed.jp is an experiment that takes this unforgettableness and turns it, whole, into a tool for learning grammar. We use outlandish, impossible English sentences as memory anchors, and we tie grammar to them. Knowledge that would vanish overnight from a bland example sentence gets strapped to the back of a giant anteater and carried home.

This article is an honest, end-to-end write-up — as the people who built the department — of the intent behind the design and the guts of the implementation. The design philosophy, written with heat; the implementation, written concretely; and "what isn't done yet," not hidden.

Memory anchors and transfer learning

At the center is a simple fact about cognition: people strongly remember information that moved them emotionally. Surprise, laughter, eeriness — any of them. Information with an emotional hook on it gets carved deeper than flat information. We repurposed this for grammar.

My husband and I are in danger because of a giant anteater.

A person who has solved this sentence learns the rule — "my husband and I" = plural subject → therefore the be-verb is are — together with the impact of the giant anteater. And later, when they meet a plural subject in a completely different problem, they recall: "ah, this is the same as that anteater one. It's are."

This is transfer learning: pulling out knowledge learned in one scene and using it in another. With a bland example ("They are students."), this recall hook is weak. You understood it the moment you learned it, but when you reach for it there's nothing to grab. An impossible English sentence builds that grip excessively strong. The heavier the anchor, the easier it is to haul back up later.

Here are real examples of the material.

• The giant anteater (be-verb): the source is a quirky line from a Nigerian scam email, "my husband was killed by a giant anteater." We tie the be-verb of a plural subject to this single sentence.
• The meat dress (passive be made of): an event where a certain artist wore a dress made of real raw meat at an awards show. "That dress is made of meat = be made of," learned together with that bizarre image.

In both, the "wait, that's a thing?!" surprise comes first. The grammar follows the surprise. Reverse the order — explain the grammar first and attach the example afterward — and it falls back into boring rote memorization. Impact first, grammar second. That order is the backbone of this department.

From clickbait headline to retention — three layers of the experience

How does a single question move through the reader? We designed the experience in three layers.

The first layer is the clickbait headline in the list. On the problem-list page, a headline like "her husband was killed by a giant anteater…?!" invites the click. Here we ignite the reader's "I'm curious." They want to know what's inside, and can't help but tap. That impulse is the first strike.

The second layer is the problem body. Beyond the tap, the giant anteater actually shows up. If you bait only with the headline and the body is bland, the reader feels let down and leaves. So we keep the impact in the problem body too, and slip the surprise in again at the moment of solving. In the instant they pick the be-verb, the anteater image stays in their head.

The third layer is the sourced trivia. After the answer, we add a sourced tidbit: "actually, this scam email is called a '419 scam'…" Here the knowledge gains depth. The reader carries home not just the grammar but "a little something they'll want to tell someone." This is where the bridge from remember to tell gets built.

Ignite with the headline, surprise in the body, add depth with the trivia. When these three layers connect cleanly, the reader leaves holding "that was fun" and "I remember it" at the same time. If even one layer is bland, the experience cools partway through. So keeping the impact alive across all three layers is what we're most careful about.

There's one more piece: the design of the entrance to sharing. When something is shared on social media, if the answer is visible in the card preview, the fun is spoiled. So the card's meta information shows the clickbait headline and hides the answer. It stops at "my husband and I are in danger because of an anteater… so, what's the be-verb?" We've structurally built in "you tap because you're curious." That snag — being unable to not tap because you're curious — is something we deliberately leave in by design.

Fun is no excuse to distort the facts

This is a passage I want to state firmly.

The more you aim for impact, the easier it is to tip into exaggeration and assertion. "If it's to make it fun, surely it's fine to embellish the story a little" — that temptation is always there. But we put up a fence against it. Fun is no excuse to distort the facts.

The trivia always comes with a source. We line up the URLs of primary information so the reader can verify it themselves. Source links are set with rel=nofollow. And we avoid flat assertions. For claims without solid proof, we add "there's a theory that…" and don't declare it. We don't generalize about races or nationalities. Even when handling scam-email material, we never write in a way that demeans a specific group. We keep it to the fun of the "quirky mistranslation" phenomenon itself.

Why be this careful? Because the more a medium pulls people in with fun, the faster inaccuracy spreads through it. Impact carries reach. If something with reach is wrong, it spreads along with the error. So fun and accuracy aren't a trade-off where you give up one — we built them into the design as a constraint that satisfies both at once. Fun, and also correct. If we can't hold both fronts, we don't ship that question.

Add data, and questions multiply

From here it's the implementation. However fine the design philosophy is, if "adding a question is painful," the volume production stops. So we poured our engineering into making the act of adding a question as light as possible.

The question data is held as vertically-arranged YAML, one block per question. Adding a new question means writing one block in a YAML file. We don't touch a single line of code. With horizontally-held CSV, every time you add a field you suffer adding a column and fixing the parser; with vertical YAML, you just add the needed key to a block. The shape of the data decides the lightness of production.

And proper nouns, URLs, and numbers are written in exactly one place (SSOT = Single Source of Truth). Even the slug of the flagship department lives in just one place in a config file. We don't write the same value twice. Change one place, and it reflects everywhere. This pays off more and more as questions grow. We can structurally crush the accident where "in just one of a hundred questions, an old value lingers."

Each question has a fixed URL (something like /topic/impossible-english/q/anteater-danger). And it outputs Quiz structured data (JSON-LD). Search engines correctly understand "this is a quiz" and show it richly in results. Add a question, and it goes onto the sitemap automatically. Add one block of data, and the URL, the structured data, and the sitemap all come along. All the creator does is write a block in YAML — that, and nothing else.

We also built the navigation flow into the data structure.

• Department hub (/topic/impossible-english/): the catch-basin that aggregates the impact department's questions.
• Cross hub (a form like /topic/impossible-english/passive-voice/): the intersection of "impact department × passive voice," the crossing point of department and grammar.

By creating the flow of "on to another question of the same grammar," a path is born: anchor (the impossible sentence) → the existing question set → retention. It's not "solve one and you're done"; your hand reaches for the next problem over.

The question-making itself we also turned into a standard procedure. Decide the material → verify the facts → generate the problem → add it to YAML → audit it. We formalized this process and also hold a mechanism for drafting multiple materials in parallel. We clearly separate the human nose for choosing material from the mechanization of production after that. The nose part stays in human hands; the volume part is left to the mechanism.

Honest limits

As a build-in-public record, we also write what isn't done.

First, the number of questions is still small. We built the mechanism so that "add data and they grow," but the crucial substance — material that is both fun and accurate — isn't yet present in sufficient quantity. The mechanism being able to withstand volume production, and volume production actually running, are two different things. Right now we're still at the stage where a shelf that just opened holds only a countable handful of questions.

And the choice of material relies on the human nose. The judgment of "this has impact" and "this works as a grammar hook" cannot be automated for now. What can be mechanically mass-produced is only the production process after the material is decided. The first step — "what to make the material" — a person picks one by one. This is the bottleneck, and at the same time the last fortress guarding this department's quality. Whether the nose should be, or even can be, replaced by a mechanism — that itself is a question with no answer yet.

Honestly, this is a department where the concept leads and the implementation and track record are chasing it. Even so, the backbone of the design — impact first, accuracy protected by sources, growth by just adding data — has been set, unwavering, from the start.

Our aim — you touch it because it's fun, you remember it because you touched it

Finally, our aim.

What we want to build is something beyond just an English-learning site. An impactful question becomes the hook for sharing. It gets shared on social media as "look at this," and companies and media want to feature it as "what a fun way to use it." The person who came via that hook gets caught by the department hub as a catch-basin. Hook them in, let them roam the hub, carry them to retention. In the future, collaboration questions with beverage or confectionery makers, too — on the data structure, we can line them up in the department by adding just one question. Cast a flagship product into an impactful sentence and make it a grammar anchor. We've left that kind of expansion open as a structure from the start.

And the most important chain is this.

You touch it because it's fun. You remember it because you touched it. You tell someone because you remember it. It spreads because you told someone. The first strike of that chain is the "impossible English sentence." Make them touch with a headline they can't help but tap, carve the grammar in along with the surprise, and send them back holding a piece of trivia they'll want to tell. We swap the starting point of this chain from boring rote memorization to an unforgettable sentence. That is the whole of this department.

If you'd like to try memorizing English grammar by strapping it to the back of a giant anteater, take a peek at english.passed.jp if you like.