desloplisted

# Deslop ## What this is A writing procedure, not a style. It exists because of one finding that survives every study in the research behind it: **model output is variance-compressed.** Models make the same choices as themselves across runs, and nearly the same choices as each other — at every level from function words to plot structure. That compression, not any particular word or dash, is what makes machine prose recognizable and boring to anyone who reads a lot of it. The compression has a cause. Human writers select *against* their own defaults under commitments that come from outside the text — what they mean, who they're for, what they're willing to be wrong about, what they refuse to do. Assistant-tuned models select *toward* their defaults, because nothing is pushing the other way. The gap is not capability. It is selection pressure. So this skill installs the pressure. It does not describe good writing and ask for it. > **Prime directive: never aim at "human style."** There is no human centroid — the human "cluster" in every study is an *absence* of clustering. Imitating human style just builds a new tight cluster. Aim at committed, situated, selective writing; reading as human is a side effect of a real author actually being present in the choices. Work top-down, always in this order: **decisions → register → surface.** Every "humanizer" works bottom-up (swap words, vary sentences) and fails, because style edits move the surface without moving a single decisio