fuck-slop

Run a mechanical scan using regex patterns targeting negative parallelism, puffery vocabulary, hedging phrases, and false ranges, then rewrite each finding by asserting its underlying claim. Re-scan the rewritten text and repeat until a pass produces zero pattern hits.

The highest-priority tell is negative parallelism ("not X but Y" and variants like "less about X than Y"). Other strong signals include uniform sentence length, em-dash density above one per 150 words, rule-of-three lists, and words like delve, tapestry, seamless, and pivotal.

Yes. The tell catalog is written as grep -Ein compatible extended regex patterns, so you can run them literally against a draft file. Sentence-length variance can be measured with tr, sed, and awk to flag uniform cadence.

Yes, each genre has distinct fatal tells and goals. Reddit forbids bold, bullets, and "Hope this helps!", while academic writing keeps formality and long sentences but cuts significance puffery like "crucial implications" and "delve".

Rewriting reintroduces slop because the same model priors that created the patterns produce paraphrased versions of them, such as turning "not just X, it's Y" into "less about X than Y". A verify loop that re-scans every rewrite until clean is required.

Avoid overcorrection: do not add fake typos, forced slang, or manufactured voice, and do not strip em dashes entirely since humans use them. In technical docs, keep legitimate "Term: definition" bullet structures and never trade precision for personality.