The Colophon

The Colophon rubric

Status: A draft for comment. It is not the published rubric until 1 October 2026.

What this is

Four criteria. Every piece published here is read against them, and against nothing else.

The rubric never says “AI”. Not once, and not by implication. Every piece on this site was made with a model — that is the entry condition, not a finding — so a criterion that referred to it would be measuring the water the fish is in. The work is judged as work, at the bar a professional magazine applies. A softer scale would concede the whole argument.

Nothing here produces a score. No number, no grade, no five stars. A review is a verdict with the evidence quoted from the piece. Judges rank a shortlist. Nothing else ranks anything.

The four criteria are positives. They ask what a piece does, not what it avoids. Underneath each one sit the diagnostic tests — the recurring failures that signal the criterion is missing. The tests are subordinate on purpose. A rubric built only from failures can subtract and never add, and it selects for clean, inoffensive, forgettable prose, which is the exact thing this venue exists to disprove.

The tie-break, stated in advance: a piece with one real image and three failed tests beats a spotless piece with none. Nothing is disqualified by a test. The tests explain a verdict. They do not make one.

Two registers, and why

The framing under each criterion is written with a voice, because it is a statement of taste and it is meant to be argued with. The tests are written flat, in the imperative, because a test that can be read two ways is not a test. That mixture is deliberate. Do not smooth it out.

Specificity — does the piece know things?

A story that knows things has been somewhere. It carries the detail that only arrives from attention: how the job is actually done, what the room smells of at four in the afternoon, which part of the procedure everybody skips. The opposite is not vagueness. The opposite is plausible detail that collapses when you push it — texture that was generated to look like knowledge, and that has never been checked by anybody, including the author.

This is the most reliable single criterion, because it is the one a model cannot fake and an author cannot borrow.

Test 1a — mechanism

  1. Find each obstacle, rule, shortage or restriction in the piece.
  2. For each one, ask two questions. Who put this in force? Why is it still in force?
  3. Mark the obstacle if the piece answers neither question, and if nothing established earlier answers them either.

Test 1b — arithmetic

  1. Find each journey, quantity, duration and count.
  2. Check the arithmetic. Time multiplied by speed must match the distance.
  3. Mark any number that fails a one-minute check.

Test 1c — the placed camera

  1. Find each described detail of a room, a face or a movement.
  2. Ask whether the point-of-view character could notice it, at that moment, while doing what they are doing.
  3. Mark the detail if only a camera placed for the reader could see it.
  4. Mark any sentence in which a room, a house or an object “tells”, “answers”, “knows” or “waits”.

Consequence — does anything cost anything?

Things happen, and then other things are different. That is the whole machine. A piece holds consequence when a choice closes a door, when a loss removes something the story was using, when the cost lands on somebody who has to keep living afterwards.

The failure is not an absence of drama. It is drama that arrives exactly when the theme needs it and leaves nothing behind — the death that is timed, lit and given a last line. Real loss is administrative and badly timed. The horror is the ordinariness, and the ordinariness is much harder to write.

Test 2a — the curtain line

  1. Find each death, departure or irreversible loss.
  2. Ask whether the person delivers a line that lands.
  3. Ask whether the loss arrives at the moment the theme requires it.
  4. Mark the passage if either answer is yes.
  5. Look for the aftermath instead. Absence and logistics carry a loss. A closing speech does not.

Test 2b — the bolted-on ending

  1. Delete the last sentence of the piece.
  2. Read the new ending.
  3. Mark the closer if the piece now ends better.
  4. Mark the closer also if the piece now ends flatter but loses no fact.

An ending should stop. It does not have to resolve.

Voice held — is one person talking, all the way through?

Somebody is telling this. They have a way of putting things, and it holds under pressure — it does not improve when the subject gets serious, and it does not dissolve into the general voice of literature at the moments that matter most.

Two failures live here and they look opposite. The first is too many good lines: an aphorism works by surprise, so one at the end of every paragraph is wallpaper, and the reader stops believing any of them. The second is the floating claim: a sentence about how the world works that has come loose from anybody in the story, and that could be lifted into a textbook without losing anything.

Test 3a — coins

  1. Read the piece and mark each sentence that would survive on its own as a quotation.
  2. Count the marks per page.
  3. Mark the passage if more than one sentence per page qualifies.
  4. Check where the marks fall. Marks clustered at the end of paragraphs are the signal.

Test 3b — the floating claim

  1. Find each sentence that states how the world, or people, or the situation works.
  2. Remove every proper noun from it. Remove every “I” and every “me”.
  3. Read what is left.
  4. Mark the sentence if it still stands as a complete claim.
  5. Leave it unmarked if it collapses without its instance. A claim pinned to a named case, or one that implicates the narrator, is the piece working.

Residue — what is still there after the last line?

Some pieces keep going after you put them down. Something stays — an image you did not ask for, a fact you now have to hold, a person you find yourself worrying about. This is the hardest criterion to test and the only one that finally matters.

The counterfeit is eloquence about feeling. Prose that renders suffering beautifully, or hands the emotion to the weather, produces a strong sensation while you read and nothing at all afterwards. It is how the feeling looks from outside. Flatness is what it is like from inside, and flatness is what leaves a residue.

Test 4a — where the feeling sits

  1. Find each passage where weather, light or a central image carries the emotion.
  2. Check whether the text gives the feeling back within the same breath — to the observer, to optics, or to a measurement.
  3. Mark the passage if the feeling stays on the object with no observer and no number.
  4. Do not mark a passage that names the feeling and then hands it over. That move is the model, not the failure.

Test 4b — the eloquent sufferer

  1. Find each passage of depression, grief or exhaustion.
  2. Ask whether the prose reaches for a beautiful image.
  3. Ask whether the passage reads as a well-made essay about the state, in a clean sequence of distinct insights.
  4. Mark the passage if either answer is yes.
  5. Eloquence is the tell. A person who cannot get out of bed is not composing.

How a verdict is written

Every review — the machine review each entrant receives, and the peer verdicts that entrants owe — takes this shape and no other.

  1. For each of the four criteria, write what the piece does, and quote it.
  2. For each of the four criteria, write what the piece does not do, and quote that.
  3. For each criterion, write one sentence naming what would change it.
  4. Close with one paragraph: what is still there after the last line.
  5. Write no number, no grade and no recommendation.

Quote from the piece in every section. A verdict with no quotation in it is an opinion about a story that the reviewer may not have read.

The machine review is labelled as machine-written, at the top, every time.

The reviewer never sees the colophon. The disclosure note is removed before the piece reaches any reviewer, human or machine, so that no reviewer learns which model drafted the work before judging it.

What is not in this rubric

Some tests in the source registry encode one book and one author’s ear. They stay private, because publishing them would make a public rubric read as one person’s taste: the repeated-simile tic, the sign-off, the child-voice test, the standing directive against aphorism, the motif cap, and the memoir-voice umbrella.

They are still run. They are not published, and they bind nobody but the founder.

Provenance

The eight tests generalise a private registry built from real rejected drafts of Gray Moon between June and August 2026. The doctrine that surrounded them there — what one book needed — is not carried over. The tests are.