← Playground一覧に戻る

Morning Edition

The recent run of a machine that writes one short story every morning — each set beside the three Aozora Bunko texts it was made from, the notes on what it took from them, its critique and its scores, in a three-column sheet that never grows longer than the screen however many mornings it holds.

Morning Edition

Nos. 01–20 2026.08.14 → 2026.08.24 20 stories

Order

Index

No. 20 Mon 2026.08.24 831 chars Craft 58/70

水をやる手

深夜二時、削除待ちの列にはまだ数百件が並んでいる。彼女はモニターの前で、規定に触れる語へ次々と下線を引いていく。〈死ね〉。〈お前の会社は潰れろ〉。〈これ買うと五キロ痩せました〉。〈○○さんの家はこのへんです〉。基準は明快で、迷う必要はほとんどない。手首の動きは机の木目のように均一になっていく。

棚の隅には、前任者が置いていった多肉植物の鉢がある。水やりの係は誰にも頼まれないまま彼女に回ってきた。去年の秋から花をつけたことはない。それでも枯れはしない。週に一度、コップの水を土にかける。それだけの関わりだった。

その夜、列の途中に一件、システムが判断を保留した投稿が挟まっていた。〈父は三年、便りをよこさない。便りを書ける場所にいないのだと、母は言わない〉。個人情報の項目には触れず、暴力的な語彙もない。ただ「収容」「拘束」といった語が離れた行に散っていて、自動判定が人の目に投げてきたのだった。

彼女はカーソルを止めた。学生の頃、詩のようなものを書いていた。書いたものは、実家の抽斗の底に何枚か眠っているはずだが、読み返したことはない。就職の面接で、その話をしたことは一度もない。

規定の欄には「政治的主張に該当するかは文脈を総合して判断する」とだけある。文脈を総合するための時間は与えられていない。次の投稿はもう列の先頭に並んでいる。カーソルは保留のボタンへ向かいかけ、そこで止まる。

指は、数百件を消してきたのと同じ角度で、いつもの位置へ吸い寄せられていく。理由欄には「関連語ヒット」とだけ入力した。手つきは、〈死ね〉を消した時と変わらなかった。投稿は列から消え、次の行が画面を占める。

作業を終えたのは明け方近くだった。彼女はコップに水を汲みに立ち、多肉植物の鉢の前で足を止めた。土は乾いていた。水をかける。土の色が変わり、光る。蕾はどこにも見当たらない。彼女はコップを流しに戻し、また椅子に座った。換気扇の音が、規則正しく回り続けている。

No. 20 · 1 of 20

Morning Edition

Every morning a program picks three works from 青空文庫 — the public-domain library of Japanese literature — reads them, and writes one short story of its own. Not a summary and not a pastiche: it takes a structure from one, a motif from another, a cadence from the third, and writes something none of them contains. Then it is critiqued against a rubric, revised if it fails, scored for how far it actually got from its sources, and mailed. By the next morning it has done it again.

This sheet is the recent run of those mornings, and it is laid out as a daily paper’s literary page: an index down the left, the story set in the middle column, and the apparatus — where it came from, what it took, what the critic said, how it scored — in the right-hand column. Each column scrolls on its own, so the whole edition sits inside one screen however many stories it holds. The masthead states which numbers and which mornings are on the sheet, because that is the part that changes.

The three columns

ColumnWhat it carries
IndexEvery story in the edition: its number, the morning it was written, its title, and a rule showing its craft score
StoryThe piece itself, set with the first line of each paragraph indented, as prose is set on paper
NotesSource texts, inspiration, critique, scores, colophon — everything about the story that is not the story

Because each column moves on its own, each one says where it is: a rule down its right edge marks how much of that column you are looking at and where that part sits in the whole of it — how far down the index you have run, how much of the story is behind you, whether the notes go on past the fold. The rule is only there when there is more than fits, so its absence means you are seeing all of it. It can be taken hold of and dragged, like the sliding index tab on a ledger.

Choose a story from the index, or step through with ← → from anywhere on the sheet. The index also takes ↑ ↓, Home and End. Order re-sorts the index by date or by craft score without losing your place; the numbers stay attached to the mornings they belong to, so a re-sorted index still reads as an archive rather than a leaderboard. On a phone the three columns become three panes with a switch above them. Selecting a story writes it into the address, so a particular morning can be linked to.

Where the numbers come from

The story is judged twice, by two things that cannot see each other, and the sheet prints both plates rather than averaging them into a single verdict.

Craft is a critique against a seven-axis rubric, out of seventy. It is a reading, not a measurement — the same faculty that wrote the story, turned on the result and told to be unkind.

AxisWhat it asks
NarrativeDoes the causality hold, and do the details planted early pay out?
InevitabilityDoes the turn arrive from the story’s own logic, or from a convenient accident?
ProseIs the sentence-level writing controlled, or padded?
JapaneseDoes it read as Japanese, or as a translation of an idea about Japanese?
OriginalityIs there a leap, or only competent assembly?
InheritanceDid it inherit from its sources rather than copy them?
ResonanceDoes anything stay after the last line?

Creativity is arithmetic, out of a hundred: character bigrams from the story are set against the bigrams of each source text. Dissimilarity is the inverse of the closest overlap with any one source, Originality is the share of the story’s bigrams that appear in none of them, and Reminiscence is how much of a single source’s texture the story reproduces. High dissimilarity with high reminiscence is the signature of borrowed phrasing, which is exactly what the critique is told to look for. These three are printed in the second ink, because nothing about them is a reading.

The two disagree often, and that is the point of showing them side by side. A story can score well on distance from its sources and poorly as a story.

How the sheet is made

The stories are not fetched at runtime. A command in the generator exports the most recent ones as a single JSON file, the file is committed beside this page, and the whole edition — every story, every note, every score bar — is rendered into HTML at build time. Nothing is drawn by script: the page works with JavaScript switched off, showing the most recent morning in full. What the script does is switch which story is shown, re-sort the index by setting a CSS order, and keep the address bar in step. Nothing about the length of the run is written into this page: re-run the export with any number and the sheet takes as many stories as it is given, since the numbering, the dates in the masthead and the ordering all follow the file.