Two Anki decks
from your Gemini Notebook sources

.card { font-family: "Hiragino Sans", sans-serif; font-size: 22px;
        text-align: center; color: #2b2b2b; background: #fffdfa; }
.word { font-size: 46px; font-weight: 600; }
.sent { font-size: 26px; line-height: 1.9; margin: 14px 0; }
.kana { font-size: 16px; color: #8a8a8a; }
img { max-width: 82%; border-radius: 10px; }

Go through the sources in my notebook and get them ready for making Japanese flashcards. Everything I build will start from sentences, so look at them with that in mind.

1. List every source in one line each, and say what kind of material it is (a worksheet from a lesson, my own speech, an article, a video transcript, and so on).
2. For each source, tell me how much of it can become spoken Japanese sentences, and how much is written-style material I should keep out.
3. List the everyday situations my sources cover, starting with the ones that come up most (at home, in a shop, at work, and so on).
4. Name three everyday situations I clearly need Japanese for that my sources barely cover (for example seeing a doctor, taking a phone call, asking for directions).

No tables. Short bullet points.

I'm going to talk casually in English about my daily life for about three minutes. My work, what I did today, what I'm into lately, the things I always end up talking about with friends and family.

Act like a close, easygoing friend. Just listen, react naturally, and ask me ONE specific follow-up question at a time to keep me talking.

Do not lecture me. Do not correct my English. Do not over-praise. Keep your questions short and specific.

I'll be using this later to build my own Japanese vocabulary, so keep me talking about things I would actually say again.

Let's start.

Now pull out the phrases I actually use, from the conversation we just had.

HOW TO CHOOSE THEM
1. Prefer anything I said twice, and anything I kept circling back to in different words.
2. Take whole phrases, not single words. Not "usually" but "I usually end up working late".
3. Always include the places where I got stuck, or explained something the long way around. Those are the phrases I am missing most.
4. Do not add anything I did not say. Common textbook phrases stay out unless they came out of my mouth.
5. Stop at 30.

OUTPUT
No preamble and no explanation. Give me one table with these three columns.

| Situation | The English phrase I used | How often I'd say it (daily / a few times a week / now and then) |

Using my sources, write Japanese sentences for me to turn into Anki cloze cards. This table is the base for my whole deck, so treat it as the most important step.

THE LEARNER
An adult who already speaks English. Upper beginner in Japanese. Knows plenty of words, but keeps getting the particles wrong and freezes when it is time to speak.

CHOOSING THE SENTENCES
1. Base each sentence on a situation, a topic or a line that appears in my sources.
2. Write sentences people actually say. No explanatory or written-style sentences.
3. Short enough to say in one breath, roughly 12 to 20 morae.
4. Every sentence should carry at least one word worth learning, because I will be cutting my word cards out of these same sentences later.
5. Do not repeat the same shape twice in a row. Spread the settings around: at home, in a shop, at the station, at work, talking with a friend.
6. Only use kanji that appear in my sources. If a kanji is not in my sources, write that word in hiragana. The front of the card shows no reading, so a single unreadable kanji makes the card unanswerable.

REMOVING THE PARTICLE, which is the part that matters most
1. Remove exactly one particle per sentence.
2. Only write sentences where one and only one particle fits the blank. If two could fit, do not use that sentence.
   Bad: 学校（　）行きます, because に, へ and まで all fit.
   Good: ごはん（　）食べます, because only を fits.
3. Spread the removed particles across を, に, で, が, は, と, から, も, へ, まで. Never use the same particle three times in a row.
4. Write the blank in Anki cloze form, like this: ごはん{{c1::を}}食べます。

THE IMAGE QUERY COLUMN, which goes straight into a Google image search
- Write it in English, two to five words. Keep it short.
- It has to be something a photo search can answer. Say the scene of the sentence in terms of people, objects and place.
- No abstract words (communication, feeling) and no grammar terms.
- Always attach a person to an action. Not waiting, but woman waiting at station.
- No brand names and no famous people.

OUTPUT
No preamble, no explanation, no citation numbers, no headings. Give me one table, with the columns in the order below. One sentence per row. Stop at 25 sentences.
IDs run from alpc_s001 upward. Reading is the complete sentence, answer included, written out in hiragana. Never use romaji.

| ID | Text | Reading | ImageQuery |
| alpc_s001 | ごはん{{c1::を}}食べてから、出かけます。 | ごはんをたべてから、でかけます。 | man eating rice bowl breakfast |
| alpc_s002 | 駅の前{{c1::で}}友だちを待ちました。 | えきのまえでともだちをまちました。 | woman waiting in front of station |

REMOVING THE WORD
1. Remove exactly one word I need to learn per sentence. Leave the particles alone.
2. Do not write sentences where the rest of the sentence gives the answer away.
3. Write the blank in Anki cloze form, like this: 駅の前で{{c1::待ちました}}。

Now pull my word cards out of the sentences you just wrote. Every word card has to sit on top of one of those sentences.

CHOOSING THE WORDS
1. Only take words that appear in the sentences you just gave me. Nothing from outside that table.
2. Take one to three words from each sentence, the ones actually worth a card of their own.
3. Skip the particles, and skip the very basic words such as する, ある, これ, いい.
4. Take each word once only. If it shows up in several sentences, keep it with the sentence that shows its meaning best.
5. In the Word column, put verbs and adjectives in their dictionary form even when the sentence conjugates them. The sentence keeps its natural conjugated form.
6. Only use kanji that appear in my sources. If a kanji is not in my sources, write that word in hiragana. The front of the card shows no reading, so a single unreadable kanji makes the card unanswerable.

THE COLUMNS
- Source is the ID of the sentence the word came from.
- Sentence is that sentence written out normally, with the missing particle filled back in and the cloze braces removed. This matters: if any {{c1:: }} is left in the Sentence column, the card will print it as text.
- SentenceKana is that same sentence in hiragana.
- ImageQuery is English, two to five words, and it has to show the meaning of THE WORD in a photo, not the scene of the sentence. No abstract words, no brand names. Always attach a person to an action.

OUTPUT
No preamble, no explanation, no citation numbers, no headings. Give me one table, with the columns in the order below. One word per row.
IDs run from alpc_w001 upward. Reading is hiragana only. Never use romaji.

| ID | Source | Word | Reading | Sentence | SentenceKana | ImageQuery |
| alpc_w001 | alpc_s001 | 出かける | でかける | ごはんを食べてから、出かけます。 | ごはんをたべてから、でかけます。 | man leaving house front door |
| alpc_w002 | alpc_s002 | 待つ | まつ | 駅の前で友だちを待ちました。 | えきのまえでともだちをまちました。 | woman waiting checking watch |

Same format, keep going with 25 more. Nothing you have already given me. Continue the IDs from where you stopped. The table only.

Look over the two tables below (the sentence table and the word table) as a Japanese teacher would, be strict about it, and then give me the corrected version.

WHAT TO CHECK
1. Is there any sentence where two or more particles could fit the blank? This matters most. Rewrite those sentences so that exactly one answer fits.
2. Would a Japanese native speaker really say it this way? Watch for odd word order, stiff phrasing, and textbook Japanese.
3. In the word table, is every word actually used in the sentence next to it?
4. In the word table, is any {{c1:: }} still left in the Sentence column? Remove it and write the sentence out normally.
5. Are the readings correct, and is any romaji creeping in?
6. Is the same word listed twice, or are two sentences built on the same pattern?
7. Is any image query an abstract word, a grammar term or a brand name? It has to be something a photo search can answer.
8. Does any kanji break the kanji rule above? Open those words up into hiragana.

HOW TO ANSWER
First, list only the rows you changed, one line each, as ID and reason. Keep it short.
Then give me back two tables with every correction applied: the sentence table, and the word table. **Do not merge them into one.**
Keep each one in exactly the same columns and the same order as the original. Include the header rows, and include every unchanged row, so that these two tables are the finished thing. No preamble and no citation numbers.
If it gets too long and cuts off, stop there. When I say "continue", give me the rest.

(paste the sentence table here)

(paste the word table here)

pbpaste > ~/Desktop/alpc_sentences.txt

Get-Clipboard | Out-File -Encoding utf8 "$HOME\Desktop\alpc_sentences.txt"

Using my sources, write Japanese sentences for practising verb forms. The card shape is the same as my particle deck.

THE LEARNER
An adult who already speaks English. Upper beginner in Japanese. Knows the words, but the right verb form does not arrive in time when speaking.

CHOOSING THE SENTENCES
1. Base each sentence on a situation or a topic from my sources.
2. Write sentences people actually say. No explanatory or written-style sentences.
3. Short enough to say in one breath, roughly 12 to 20 morae.
4. One verb per sentence. Never two.
5. Only use kanji that appear in my sources. If a kanji is not in my sources, write that word in hiragana. The front of the card shows no reading, so a single unreadable kanji makes the card unanswerable.

REMOVING THE VERB, which is the part that matters most
1. Remove exactly one verb per sentence. Leave the particles alone.
2. Only write sentences where one and only one verb fits. The words around the blank have to decide it.
   Bad: 友だちを（　） because 待った, 見た and 呼んだ all fit.
   Good: 駅の前で30分も友だちを（　） because only 待ちました fits.
3. Remove the conjugated form as it stands. Do not convert it back to the dictionary form.
4. Put the dictionary form in as the hint, using Anki's hint syntax.
   Example: 駅の前で30分も友だちを{{c1::待ちました::待つ}}。
5. Spread the forms around: ます, past, て form, negative, ている, たい. Never the same form three times in a row.

THE IMAGE QUERY COLUMN
- English, two to five words, describing someone performing the action.
- No abstract words, no grammar terms, no brand names.

OUTPUT
No preamble, no explanation, no citation numbers, no headings. One table, columns in the order below. One sentence per row. Stop at 25.
IDs run from alpc_v001 upward. Reading is the complete sentence with the answer in it, written out in hiragana. Never romaji.

| ID | Text | Reading | ImageQuery |
| alpc_v001 | 駅の前で30分も友だちを{{c1::待ちました::待つ}}。 | えきのまえでさんじゅっぷんもともだちをまちました。 | woman waiting outside station checking phone |
| alpc_v002 | 牛乳を買うのを{{c1::忘れて::忘れる}}しまいました。 | ぎゅうにゅうをかうのをわすれてしまいました。 | man looking at empty fridge shelf |

Using my sources, write Japanese sentences for practising nouns. The card shape is the same as my particle deck.

THE LEARNER
An adult who already speaks English. Upper beginner in Japanese. Recognises nouns when hearing them, but cannot produce them.

CHOOSING THE SENTENCES
1. Base each sentence on a situation or a topic from my sources.
2. Write sentences people actually say.
3. Short enough to say in one breath, roughly 12 to 20 morae.
4. Spread the settings around: at home, in a shop, at the station, at work, with a friend.
5. Only use kanji that appear in my sources. If a kanji is not in my sources, write that word in hiragana. The front of the card shows no reading, so a single unreadable kanji makes the card unanswerable.

REMOVING THE NOUN
1. Remove exactly one noun per sentence. Leave the particles and the verb alone.
2. Only write sentences where one and only one noun fits. The words around the blank have to decide it.
   Bad: （　）を買いました because anything fits.
   Good: 毎朝、（　）から牛乳を出します because only 冷蔵庫 fits.
3. No hint. Use the plain form {{c1::冷蔵庫}}, not the hint syntax.
4. **Only choose nouns a photo can show.** Objects, places, people, food. Never feelings, situations or ideas.

THE IMAGE QUERY COLUMN
- English, two to five words, and it has to return a photo of **the noun itself, not the scene**.
- For 冷蔵庫, write refrigerator in small kitchen. A photo where the fridge is the subject, not a kitchen view.
- No abstract words, no grammar terms, no brand names.

OUTPUT
No preamble, no explanation, no citation numbers, no headings. One table, columns in the order below. One sentence per row. Stop at 25.
IDs run from alpc_n001 upward. Reading is the complete sentence with the answer in it, written out in hiragana. Never romaji.

| ID | Text | Reading | ImageQuery |
| alpc_n001 | 毎朝、{{c1::冷蔵庫}}から牛乳を出します。 | まいあさ、れいぞうこからぎゅうにゅうをだします。 | refrigerator in small kitchen |
| alpc_n002 | {{c1::傘}}を電車に忘れてしまいました。 | かさをでんしゃにわすれてしまいました。 | folded umbrella left on train seat |