Q: How are slop managed? What about differentiating writing styles?

[TomLucidor]

Some of the example I have are in other lists

• Prose slop Q: How is this different from everything else? hardikpandya/stop-slop#51
• Code slop Comparison with other tooling peteromallet/desloppify#110 (comment)
• (not gonna write about design slop since they are less language-like)
• Also for "humanizers"
  • https://github.com/blader/humanizer
  • https://github.com/lynote-ai/humanize-text + https://github.com/lynote-ai/best-humanizer-handbook
  • https://github.com/AIScientists-Dev/academic-humanizer
  • https://github.com/anasu1/text-humanizer
  • https://github.com/Anbeeld/WRITING.md
  • https://github.com/harshaneel/humanize
  • https://github.com/lynote-ai/humanize-text-skill

The major idea here is that if a person strongly leverage AI for "writing to think", it needs preemptive work to stop generic-ness from bleeding over.

[ohad6k]

Sorry this sat so long, it deserved a real answer.

Voice is the layer emulo is weakest at, and I would rather say that than pretend otherwise. The profile can tell an agent how I decide. It is a lot worse at telling an agent how I sound.

That maps onto your question directly. Anti-slop rules are mostly about voice, and voice is the thing raw session logs carry worst, because nobody performs in a terminal. The humanizer projects you linked work on published prose, which is the opposite raw material: performed, edited, voice-heavy.

Where I think they stack is the rejection half. Sessions carry what you reject very well, because rejection is the one thing people write out in full while working. "no, cut that", "too generic", "say it plainly" show up constantly. So the honest version is a reject-list mined from sessions feeding a voice tool, rather than emulo trying to own voice itself.

0.6.2 has emulo verify, which pulls every quote out of a drafted profile and searches your sessions for it, exiting non-zero on anything it cannot find. That is the closest thing to a slop check in there today, and it only catches invented evidence. It does nothing about generic prose.

Not claiming it handles your question. Leaving this open.