The rubric, in full.

Two questions, in order

Is it excellent fiction? Then: is the account of how it was made clear, honest, and yours?

Fiction carries 80 points. The process account carries 20. A strong process account cannot rescue weak fiction, and strong fiction cannot win with a dishonest or unsupported account.

We do not judge how much of the text a machine generated, where a sentence came from, the subject (a work gets no credit for being about AI), or how technical the process was. We do not use AI-detection tools, and we never run the work through an AI model.

Stages

Six stages, from screen to final panel.

1. Administrative screen. Completeness, eligibility, and file safety. Not a judgment of quality. A person confirms every failure.
2. First read. Independent readers decide whether the work earns a full read. They see the work only: no author identity, no process materials.
3. Full read and fiction score. Advancing works are read completely and scored out of 80.
4. Process review. The strongest works are assessed on their process dossiers and scored out of 20.
5. Shortlist. The panel chooses finalists, with written reasons for each.
6. Final panel. Final judges read every finalist in full and choose the winners.

The first read

Does this earn a full read?

The first read is a floor, not a ranking. It sets aside work that isn’t ready; it doesn’t order the work that is. There is no quota. Readers look for three things:

• Control. Sentences do what they intend. The register holds. The prose doesn’t slide into stock phrasing or filler.
• Pressure. By the end of what they’ve read, the work has made a claim on the reader: a stake, a question, a voice worth following.
• Distinctiveness. It sounds like this work, not the average of all work.

Each reader decides Advance, Discuss, or Decline, with written reasons and a confidence level, without seeing anyone else’s verdict. Readers who agree settle it. When they disagree, or a decline is low-confidence, another independent reader reads the work and must choose Advance or Decline.

Short Fiction is read whole at this stage.

Fiction score · 80 points

Impact and originality.

Readers read the complete work and score it on two criteria, with written reasons for each.

• Impact (50). The work’s effect on the reader, through its prose, character, structure, pacing, and control of form.
• Originality (30). Freshness, ambition, risk, and a voice of its own.

If readers’ fiction scores for a work are far apart, another reader scores it independently. The outlying score and its reasons stay in the record for the panel.

Made with AI is expected. Unedited isn’t. Unrevised machine habits (a pattern repeated until it’s the loudest thing on the page, stock phrasing, reflexive structures) are craft failures like any other, and score as such. A deliberate machine voice, such as a character who talks like a machine or a style you chose and control, is judged on whether it works.

Process score · 20 points

Clarity and honesty; direction.

Process review asks whether the account of how the work was made is clear, honest, and authored. It doesn’t audit technology or reward a larger or smaller human share. A plain-language account with no logs or repository can earn full marks. The dossier guide shows what to include.

• Clarity and honesty (10). Can a reader understand what was done, and does the account hold up against what you submitted?
• Direction (10). Did your choices shape the work: what you asked for, kept, rejected, and redirected?

Curating generated text is direction and counts fully. Revision is one form of direction, not a separate requirement.

Two pass/fail gates. The dossier must show that a machine took part in making the work; if it’s unclear, the reviewer asks you first. And sources, permissions, and the privacy of real people must be respected. A work needs at least 8 of 20 process points to become a finalist.

Shortlist and final panel

Scores order the discussion. People decide.

The panel builds the shortlist from works that cleared process review. The combined score orders the discussion but doesn’t decide it: the panel writes reasons for every finalist, and for any work passed over despite a higher score.

Every final judge reads every finalist in full and records their own score and ranking before seeing earlier scores or anyone else’s view. Then the panel deliberates. Consensus is preferred; if it fails, a ranked-choice vote decides. The Grand Award can go to a finalist in either category. The Galatea Citation can go to any work that reached process review, if its method is genuinely inventive.

Readers and judges may not enter the cycle they judge.