humanizer's guardrails keep my system from over-correcting. thank you

[odinfree]

humanizer's dna has been in my personal writing system for months, merged with stop-slop and a few siblings into an obsidian rule set that runs before anything ships.

evaluated it against all its sources this week. the pieces i took back from yours today. the anti-compression rule (rewrite, don't delete), the cluster principle for detection (a single em dash means nothing), and the aphorism-formula catalog. anti-slop tools die by false positives and yours is the source that took that seriously.

if you want any of my additions back, crypto-specific tells or regression cases for the rules themselves, ask and i'll paste them here.

oli (@odin_free)

[blader]

Thank you for this unusually thoughtful feedback. It is especially useful to hear that the anti-compression rule, cluster principle, and false-positive restraint hold up in a real writing system. If you want to share focused regression cases here, they would be welcome. I am leaving this thread open.

[odinfree]

appreciate you reading it that closely. the cases most relevant to your three rules, in the format my suite uses. a rewrite passes when it follows the profile and preserves every supported claim.

literal inanimate subjects (false-positive restraint)
draft: "The compiler reports a missing import. The report shows the failed test."
profile: documentation.
pass: keep both subjects, they perform ordinary literal actions.
fail: invent a human actor or force passive voice.

keep the term, gloss it once (anti-compression)
draft: "The API uses an idempotency key, a request token that prevents the server from charging the same request twice."
profile: technical explanation for a general reader.
pass: keep the term and its plain gloss, repeat the same term later if the reader must track it.
fail: synonym-cycle it or delete the gloss to shorten the sentence.

rebuttal of a live belief (over-correction guard)
draft: "the quantum risk isn't the math. it's the migration."
profile: short social post where the audience actually believes the math is the risk.
pass: keep the negation, it names and rejects a belief the readers hold.
fail: flatten it to "the quantum risk is the migration" and delete the payload.

one deliberate quotable (over-correction guard)
draft: launch copy with one strong hook line and two paragraphs of plain claims.
pass: keep exactly one deliberate quotable.
fail: cut the only hook because it sounds like a pull-quote.

the full suite is 13 cases and they run after every rule edit: https://github.com/welttowelt/stop-slop-refined/blob/main/references/examples.md