Jev Moderation

A contextual brand-suitability scorer that judges each feed item against an advertiser's own versioned policy, returning a category, an exclusion flag, and a confidence for routing borderline items to review.

Cloudflare Worker that asks Jev on Workers AI for literal profanity in text or usernames and, for usernames, a second Noul for phonetic or look-alike disguises. The threshold and max policy live in Worker code, with OpenAPI docs generated via chanfana and Hono.

A bring-your-own-key Chrome extension that asks Jev one slop/not_slop Choice per X or LinkedIn post as it scrolls into view, then blurs and stamps anything at or above a configurable threshold. Scrolling never blocks, and each post is scored once from a local cache.

Chinese-language research project for multimodal social-risk moderation: a local Qwen3-VL-32B with language-side LoRA emits structured evidence JSON, and a fixed Jev policy turns that evidence into risk bands, with frozen test leaderboards and pure-Jev ablations.

Sean Brownell scores sample call transcripts in a harness, including a case where the agent reads out card details, and compares that pass with a frontier-model estimate.

A link-and-discussion site run by Tom Nick. Every post, comment, and username is classified by the TypeSafe Jev API before it can be published.

Tom Nick's write-up before jevdit.com: irony detection on 100 German soccer tweets, with Jev 1.13.0 next to Opus, Astra, and Laya.

A no-login page that scores pasted prose sentence by sentence with Jev, using a 0–4 style rubric, and highlights only the sentences that cross its threshold.

Tom Nick's Show HN for a discussion site where Jev moderates every post, with a link to the irony-detection write-up.

A self-hosted Discord bot that classifies messages with Jev in real time, deletes spam and scam links, escalates repeat offenses through warnings to timeouts, and profiles members from message history.

Private TypeScript proof of concept that flags Spanish profanity with Jev Boolean and Score questions and replaces the matching word intervals with a beep via ffmpeg, keeping the original duration.

A repeatability study: an 8-question moderation rubric run 15 times per condition shows a mean probability standard deviation of 0.0098, and a 0.60 uncertainty gate lifts agreement to 99.2%.

A plain-English guide to Choice, Score, and Noul with Python and curl examples, confidence-gated branching for support, moderation, and fintech, plus a skeptical read of the benchmarks.

Screen every message going into and out of an LLM app: a Noul describes possible hazards, a Score rates severity, and code thresholds the probabilities.