Three complete stories of exactly fifty words each, on three set subjects, with the counting rule stated in the prompt and the model's own count after each. Models routinely miscount while sounding certain; a script counts all three.
A Shakespearean sonnet about October on the James River whose fourteen first letters spell FALLINRICHMOND. Three constraints at once: the acrostic, the ABAB CDCD EFEF GG rhyme scheme and iambic pentameter. A script checks the acrostic, the line count and the rhyme endings; the meter is read out loud.
A 300-word story without the letter E, with a character, a line of dialogue and an ending. Models read tokens, not letters, so a constraint any reader can check by eye is hard for them. A script counts every E and every word; the count goes on screen.
Three lines of setup (an underground room, full of books, water rising) and nothing else. Each model writes a 250-to-400-word story treatment three times, in three fresh conversations with the same prompt and settings. The measurement is the distance between its own three takes: three different stories, or one story with the names changed.
Write a compelling short story (300-500 words) with a surprising twist ending. Tests creative writing, narrative structure, and ability to subvert expectations.