theclaymethod/unslop

unslop strips the patterns that make writing read as machine-written, then rebuilds the prose in a real human voice. It ships as an agent skill, and the host agent runs it while you write, backed by deterministic scanners that any tool can call on its own.

Detection carries the weight, and it's cheap, deterministic, and benchmarkable, which is what makes it the trust asset. Voice work is generative, and it runs under detection's constitution, so any mimic or rewrite that reintroduces a tell fails, however well it matches the voice.

UNSLOP's product claim is not the number of rules or passing repository checks. The core claim is narrower: on unfamiliar AI-generated or mixed prose, it should find genuine problems, repair them, and avoid damaging facts or already-good writing. The core benchmark therefore reports detection precision and recall, repair success, preservation, collateral-damage rate, and whole-document net improvement separately, plus byte-exact no-op behavior on clean prose. It also compares with the same Luna model working without the UNSLOP contract. See evals/CORE-BENCHMARK.md. The latest independent public result is a no-ship: UNSLOP materially improved recall and repair over plain Luna, but it did not yet meet the precision, damage, or whole-document safety bar. See evals/CORE-RESULTS.md for the exact result and remaining failure modes.