Slop score

A blunt but repeatable eval for prose: count the tells per thousand words. Tells include hedged conclusions, emoji-fronted lists, "delve", "it's important to note", triadic sentence padding, and both-sides closings. Lower is better; zero is not the goal, calibration is.

Protocol

Run the same source material through the pipeline under test, have a grader (human or model with the rubric) mark each tell with a span, divide by word count. Report the score alongside two marked examples so the number stays auditable. A pipeline steered with contrast pairs should cut its score roughly in half on the first pass; if it does not, the pair library is mislabeled.

See also

• Contrast pairs