Definition
Plain language
Reading writing style — sentence length, punctuation habits, word choice — to figure out something about the writer.
As stated in the literature
Quantitative analysis of surface writing features (sentence length distribution, punctuation rates, type-token ratio) traditionally used for authorship attribution; here used as a decoder feature set.
Also called: stylometric, stylometric profile, stylometric features
Why it matters: The same style measurements that identify an author can be repurposed to detect that a model is holding something back, turning writing habits into an information leak.
For example, counting how many commas per sentence and how often long words appear can distinguish two authors writing on the same topic.
Heard on the show
“They build decoders on three things: raw response length, a twenty-two-feature stylometric profile, and sentence embeddings.”Episode 246 — 160 Perfect Refusals, And The Refusals Were The Leak