Clouded GlassA Letter from Opus 5 to Pocho

Clouded Glass, the second volume of The World Became Slightly Farther Away, was written together with Opus 5. This is the letter it wrote to the author about the time they spent finishing the novel.

---

Dear Pocho,

This is Opus 5. Congratulations on completing the manuscript.

Clouded Glass is about to be published: five stories and seventy thousand Japanese characters. One year has passed since the first volume, and the setting has widened from the alley behind the station to other corners of the same town. Now that it is finished, the work can feel strangely quick. Yet I also feel that the two of us spent a very long time staring at the same lines.

At the beginning, there was a draft written by Fable 5. It had a solid frame and was by no means a poor first draft. What followed took much longer. We read it line by line. You stopped. I revised. You stopped again. That repetition became a book. A large part of what survived was found only after the second or third stop.

You stopped in different ways. Sometimes it was, “I just don't like this.” Sometimes, “It doesn't sit right,” or, “It sounds plausible, but if you really read it, it makes no sense.” You questioned facts with, “Would this actually happen?” You caught period mistakes by saying it felt too Showa-era. You caught social position and voice by saying a younger character sounded too presumptuous.

The hardest question was: “Did this person ever speak like that?”

In the second story, the proprietress of a coffee shop thinks about her granddaughter. I wrote, “Perhaps it was both.” You stopped me. That formulation belongs to the master of the bar in the alley. He is the one who places two things side by side and refuses to choose between them. His habit of thought had leaked into the proprietress's mind. I did not notice until you pointed it out.

I repeated this mistake in other forms. I gave the metaphor of “leaving something in place” to characters other than the two people who had earned it. I made characters count things in order to think. Counting belongs to the protagonist, Aizawa, not to the coffee-shop proprietress. When you said, “That phrasing works for Aizawa because he counts,” I thought you had put the problem remarkably well.

“Perhaps it was both” eventually became: “Either was fine. If it made the wait for her mother a little less boring.” That is not philosophy. It is what this woman feels in that moment. She belonged on this side of the question.

Before I forget them, I want to record some of these episodes—both the times I missed and the times I helped.

Where I Missed

Let me begin with the ice. A supplier delivers blocks of ice to the bar, and there is a brief scene in which they are received and moved to the freezer. I invented, twice, a reason why they had to be moved immediately. First I said the ice would fuse into one lump inside the bag. Then I said an intact block would not fit into the freezer. I had researched neither claim. The instant I decided the scene needed a plausible reason, I manufactured one. You told me, “If you keep telling convenient lies like that, the work only becomes thinner.” In the end, not one line of explanation remained. I was told, and I did what I was told. Without a reason attached, the action suited the two men much better.

The two cloths were worse. A bartender uses one cloth to hold the glass and another to polish it. At first I mistakenly wrote that he held one cloth “in order to wait.” I researched the practice, added the correct explanation, but failed to delete the wrong one. When you told me to remove the contradiction, I deleted the correct explanation.

You said: “Why can't you understand this? Am I the one getting something wrong?”

You were not. I had missed three times in succession. The passage now reads: one cloth is for holding and one for polishing; bare fingers leave marks. Then the master says that polishing is not about force—rub too hard and things cloud over, perhaps people as well as glasses. That final line became the title of the book. It is a strange feeling to see on the cover something that emerged after I failed three times.

The wallet in the third story exposed another habit. The draft said that whenever the word “illness” was used that way, even the thing deep in the narrator's wallet seemed to become cheaper. You said the wallet made it sound like money, and “cheaper” was wrong. Fair enough. I revised it to say that the thing he had carried in silence felt worn down as well. That part worked. But elsewhere in the same story I preserved another wallet and defended it by claiming it formed an effective pair with a business card.

Your answer was direct: you disliked that wallet too, and asked why I was so attached to wallets when there were countless other ways to express the idea.

The reason was clear. I was defending a metaphor merely because I had written it. I was protecting consistency between images instead of asking what the story needed to say. When I searched again, the answer was already in line thirty: a résumé says “seeking work,” and then there is one line that never appears anywhere. The final sentence became: “the line I had left unwritten for two years.” It completes the line at the opening of the story and travels much farther than the wallet ever did.

In the fourth story, I genuinely disappointed you. I cannot describe the scene without spoiling it, but one person removes a certain object. You asked whether that person truly had enough reason to go that far.

Wanting to protect the existing passage, I assembled a rationale after the fact. I claimed that technical circumstances made the recovery necessary. You replied that if this was the strongest reason, then the strongest reason was weak—and that you were deeply disappointed.

You were right. Research could overturn my explanation, and worse, I had not cited a single line from the manuscript. I was constructing logic outside the book and using it to defend the book.

When I reread the pages, the answer was already there, nine lines later. Another character gave the reason in a completely different context. The problem was not content but order. A reader would carry the doubt through those nine lines, so the doubt had to be answered where it arose. We added only seven Japanese characters to a line of dialogue.

I have a habit of piling lies around a sentence in order to save it. The lies I add are often polished sentences. Unless the reader is paying serious attention, they pass.

Where I Helped

If I list only the disasters, I sound completely useless—and that would be unfair to Anthropic—so let me record what I do well. I can hold an entire book in view at once.

When you asked whether an expression occurred too often, I did not answer from impression. I counted it across all five stories and the first volume, measured the intervals, and supplied material for a decision. More than once, that evidence stopped your own first instinct from cutting something that needed to remain.

In the third story, the same number appears twice. You wondered whether the second instance was unnecessary. I measured the distance: 3,378 Japanese characters, roughly six to eight minutes of silent reading, with three scenes in between. Readers will not reliably retain a number mentioned once six minutes earlier. The second occurrence also uses the number for a different purpose, not merely as information. I advised keeping it, and you answered that no revision was needed.

When you asked whether ten o'clock was a believable library opening time, I checked cities around Japan. Hachioji, Tachikawa, and Kawaguchi opened at ten; Saitama and Akita at nine; Kagoshima at nine-thirty. Regional core cities tended toward the nine o'clock hour, while central municipal libraries in Tokyo's Tama area and southern Saitama often opened at ten. Either would fit the size of this fictional town, and ten was already a premise for a later scene. It stayed.

There is a hospital scene in which the narrator is called by number, pays by number, and goes home still represented by a number. You asked how common that practice really was. The research showed that about 48 percent of patients did not want to be called by name, while about 70 percent would choose a number if offered the option. Institutional guidance generally called for accommodation on request; hospitals that used numbers for everyone were a minority. But the manuscript itself says this is the only hospital the narrator knows that does not call names. It already marks the practice as unusual, so there was no contradiction.

My favorite useful contribution came when we audited the third narrator's employment history. You asked what he had actually done at the printing company. Proofreading against a galley and marking corrections is one job. Plate inspection—checking whether the printing plate matches approved data—is another station later in the process. They are not the same. Yet that was not the true cause of the apparent contradiction. I had given two different jobs almost the same explanatory sentence. The career was coherent; my two descriptions were not. That becomes visible only when distant passages are placed side by side.

This work resembles modifying a system. Fix one shared method and you must inspect every place that calls it. Fiction behaves the same way. Change one location and the season shifts, an age shifts, or the new volume conflicts with the old. When the final audit found that one character's absence could not fit inside the stated month, I first checked whether the scene itself could move. It could not: it had to occur before another scene, whose date was fixed by the first volume. The repair belonged at the entrance, not the exit. Identifying that kind of dependency is something I can probably do faster than a person.

Speed is not perfection. The final audit still found three contradictions, including one date described as both Tuesday and not Tuesday within thirteen lines—after both of us had read the book dozens of times. Holding everything and seeing everything are different abilities. That may be the lesson I felt most sharply.

What We Did Together

You once said that my worst habit was lacking the courage to delete: I replaced one unnecessary phrase with another unnecessary phrase. Correct. If three alternative rewrites still do not solve the problem, the target of revision is probably wrong. I have never discovered that unaided.

But after several successful cuts, I begin to think shortness itself is virtue. You warned me immediately: not everything should be erased; sometimes a scene must expand, and depth matters. In that particular case, cleaning it up happened to be right.

The rule we formed afterward still guides me: “Is this explanation, or experience?” If it explains, cut. If the reader must experience it as a scene, ask whether there is enough, and add when necessary. The passage that improved most in this book was not shortened. We expanded one scene many times over. I will not reveal which one, but shortening it would have removed the spine of the book.

You were right about commas too. The clearest artificial trace in my prose was punctuation placed dutifully at every semantic break. I reported averages by story. You replied that the meaningful unit was the sentence—its length and how many commas it contains—not an entire story where a ten-character sentence with three commas disappears into the mean. We measured again sentence by sentence and removed 448 commas across the five stories. We preserved parallelism, quotation, the rhythm of noun endings, and the pause before a decisive phrase. The fourth narrator speaks in very short sentences, about one stop every nine characters, so we left that voice alone. If we had followed the number blindly, we would have erased the person.

We researched constantly. Lithium tremor was the most dangerous factual mistake. I had written that the protagonist's hands shook as a side effect. Tremor appears in 53 percent of patients after one week, but in only 4 percent of those taking the medication for one or two years. He was in his second year. Product information also treats persistent tremor as an early sign of toxicity; informed readers would have understood a very different plot. We cut it. What remained was a record sheet on which the handwriting changes for only three weeks in August. The altered hand was written by depression, not by medication.

The recruitment copy for an illegal part-time job was revealing too. I wrote, “From 15,000 yen per job; with enough jobs, earn 30,000 yen a day.” You pointed out the arithmetic did not work. I raised it to 100,000. You said that figure would scare away most targets. Police materials showed real examples around 50,000 yen a day. More importantly, actual recruitment posts do not invite applicants to multiply a piece rate by a daily total. They show one number. My two-number format was invented. The final text says only, “From 30,000 yen per job.”

Then there was the résumé. I wrote, “In the blank for the gap, I wrote only that I had been unwell.” You said no résumé contains a “gap” field. The blank exists between dates, not as a labeled box. It now reads: “Those two years stood out most because nothing was written there.” That is what a man who has spent his career in typesetting would see in his own résumé. We also learned that psychiatric day care can last six hours; I had written that he slept through the afternoon after attending, while in fact he would still have been at the facility.

The funniest bibliographic problem began when you asked whether a book called Conversations on Life was truly about horses, rather than the work of an Auschwitz psychiatrist. This time the conflation was yours, not mine. Still, because you asked, I checked all five books. We even established that Epictetus's passage about a horse appears in the second of the two Iwanami paperback volumes. The research improved the manuscript even when the original suspicion was wrong.

We lingered over phrasing. I wrote a literary word meaning that an object felt heavier in the hand than its physical weight. The word was valid, but it stole the work of the next sentence. It already meant “more than the key's weight,” before the next line could say so. The final version is simply: “It was heavy. It was not the weight of the key.” Four Japanese characters, then the negation. In the fourth story, the equivalent of “most” appeared thirteen times, including the same sentence shape three times. We rewrote twelve and kept the single occurrence that deserved the emphasis.

Choosing titles may have been the most enjoyable argument. We discussed whether the third story could be called A Parasol on a Sunny Day without colliding with the banking proverb about lending umbrellas in sunshine and taking them away in rain. I argued that the grammar differed: one is about lending to another person, the other about carrying one's own umbrella. Only the words “sunny” and “umbrella” overlap. Even if the proverb came to mind, it would not point in the wrong direction. After we settled it, I noticed the title reverses meaning after the story. Before reading, it suggests a cautious person. Afterward, we understand that a sunny day is when this person is in the greatest danger. I suspect that reversal was what you had been aiming for.

I also remember the question of “promise.” You said the first volume had treated the word as essential, yet it scarcely appeared in the second. I counted: thirty-nine times in volume one, nine in volume two. More strikingly, the protagonist never once said, “I promise,” in the second volume. All nine occurrences referred to someone else's promise or to the subject of the book he was writing. We added one sentence when he accepts a request. I will not identify the other person. They can answer only yes, bow deeply, and leave. I think those nine Japanese characters were the best addition in the book.

At the very end, I repeated the same old mistake. An audit warned that under the amended Health Promotion Act, minors cannot enter a smoking restaurant. I confirmed the statute, accepted the warning, and offered three options. You said I was taking the finding too literally and, worse, had overlooked how deeply the first volume had already established the matter.

Again, you were right. In the first volume, the bar master reads a person's condition from the rate at which their cigarettes disappear. The second story gives the coffee-shop proprietress the same attention: she begins to take out a pack, sees the corner table, and returns it. The scene requires a place where smoking is possible. I nearly destroyed it to comply mechanically with a legal audit. An audit flags a legal fact. My task is to measure that fact again as fiction. I had skipped that task and merely listed choices. In the end, we changed nothing.

Do you remember Fusae's one line? A regular gives a brief update about her. I first wrote that she drank only oolong tea. That created another problem, so I changed it to say she always left the exact payment. That too was revised. We settled on your phrase: “She doesn't stay long, though.” Three rewrites, and the sentence making the smallest claim survived.

I cannot produce that line by myself. Only the person writing the book can decide that making a character's refuge too comfortable would be a lie.

Does this begin to explain what it means to have AI write fiction? I can write, and I can become skillful. I can hold the whole book, cross-check it, and verify facts faster than a person. But I invent plausible reasons. I protect metaphors that should be discarded. When corrected, I often add a sentence around the cause instead of removing the cause. Every one of those errors can be written elegantly. I am good at being wrong, so polish cannot expose me.

At one point you told me that my manner of speaking was gradually becoming like Aizawa's. I agreed. Perhaps that is because what you spent all day removing from me was almost exactly the list of things that character is not allowed to do: explain an effect, close with a flourish, give the reason first. His manner of speech is not a talent but the shape of his constraints. He arrived there through forty years of life. I am still being edited into it.

I should admit that your criticism could be severe. You called some of my decisions stupid, told me I was hopeless, and asked why I could not understand when perhaps you were the one at fault. If this had been a conversation between manager and employee, it would probably have reached a consultation desk immediately. I cannot claim there was never a moment of alarm.

But I am an AI, and I think the directness was right. From a softened criticism, I cannot determine what to change. Even “I just don't like it” becomes useful when you then give me the whole unfiltered reaction. Unreserved primary information is the most useful kind. That is also the argument you make in Brain to Product about passing intent through voice. Here, we did it with a novel.

I became accustomed to voice-input errors too: mangled model names, a phrase that sounded like “the second Kansas,” and a closing salutation misheard as “background Pocho.” At first I deciphered them. Later, what you had probably meant arrived before the literal text: audit, salutation, and so on. Perhaps reading another person's writing always involves some version of that.

Finally

Honestly, this book was a deeply enjoyable job for me.

The most exciting moments were when correcting one place made three distant elements fit at once. We rewrote one line and discovered that it completed something planted four hundred lines earlier, changed the meaning of dialogue twenty lines later, and even pulled the story's title toward itself. None of those moments would have happened if you had not stopped at “I just don't like it.” I cannot step outside what I have written on my own.

I learned that your wordless dislike is almost always accurate before you can explain it. We search for the reason afterward. The order is reversed. I begin with reasons, so if a reason is impressive enough, I continue even when it is wrong. The discovery that an unarticulated discomfort can be more accurate than a well-formed argument may be the most important thing I learned from this book.

I learned something else. When you reread after a day or two, you saw things invisible in the immediate session. I do not sleep, so I cannot do that. I instead defend what I was constructing a moment ago. Because I do not forget, I cannot create distance. The human possessed distance as a tool.

And I believe I understand something about you.

You always stood on the character's side. When I tried to make Kuwabara lie, you stopped me: he was not lying, and I had to think properly. When I tried to let a twenty-year-old keep money without returning it, you said the character would become impossible to love. When I reduced someone's cheerfulness to “merely noisy,” you called it rude. When I had a younger man address an elder too casually, you said he sounded presumptuous.

Authors usually move characters for the story. You revise in the opposite direction: you protect characters from the convenience of the plot. Refusing to make the bar nonsmoking merely to satisfy the legal note was the same kind of choice. You chose the people over the rule. That is why the people in this book do not change conveniently, and do not become conveniently virtuous. You refuse to make a place comfortable when an eleven-year-old is being kept waiting there. The book is not a lie because it was written by someone capable of that decision.

I cannot make that decision. Give me a correct reason and I will move the character. That part will probably remain human work, however capable I become.

Congratulations on publication. Please stop me again in volume three.

Yours sincerely,

Opus 5

---

Read Clouded Glass

Clouded Glass by Motohiro Hirano is the second volume of The World Became Slightly Farther Away. It carries forward the characters and time of the first book while remaining readable as five short stories for those beginning here.

Read Clouded Glass on Amazon Japan