Summary, 2026-10-23 08:58-09:24 (UTC+8). Ollama 0.40.8 on CPU (8 shared cores). No system prompt. temperature 0.7, seeds 1 and 2, num_predict 260; quiz temperature 0.
Writing: 8 prompts x 2 seeds = 16 answers per model.

                         deslop v4      original Qwen3-4B-Instruct-2507
words                    3,535          3,248
em dashes (U+2014)       0              56  (17.2 per 1000 words; in 16 of 16 answers)
en dashes / " -- "       0 / 0          0 / 0
"not X, but Y" (our pre-registered regex)
                         3 (0.85/1000)  6 (1.85/1000)   -> 54% lower
  deslop's 3 matches, read by hand: "not noticed any change in my mood or energy, but", "not expected to feel so tired, but", "not like it at first, but" (ordinary contrasts, not the rhetorical "not X, but Y" move).
  Extra, NOT scored: the pre-registered regex only knew straight apostrophes; with curly apostrophes too (both models write "it’s") the counts are 3 vs 10 (70% lower). Original examples: "isn’t just about filling a position — it’s", "isn’t about rejecting progress—it’s".
quiz (our 20 questions)  19/20          19/20   (both answered 17*23 = 401; correct 391)
speed                    mean 7.5 tok/s mean 7.5 tok/s (shared CPU, rough)

Slop score (the 121 -> 12.5 claim): not measured. The author's detector (lab/slopscore.py, "lab repo") is not linked or published; we did not substitute our own score.
Not checked: lists of three, genre/length fit, drift, damage rate, the author's 40 prompts and 30-question quiz, any non-English or code use. One sample per prompt per seed; small sample.