LinkedIn NoSlop

It is a Manifest V3 Chrome extension that finds feed posts on linkedin.com/feed, extracts only the post text, and blurs the ones Jev judges low-value. Blurred posts show the reason and a one-click Show post button; slop means your own definition, not a claim a post was written by AI.

The background worker sends the post text to TypeSafe/Jev with a rubric of narrow yes/no and 0–2 questions. Plain code in decide.js combines the answers into keep, filter, or uncertain; only filter blurs, and anything uncertain, incomplete, or failed stays visible.

Get a TypeSafe API key from the dashboard, open chrome://extensions, enable Developer mode, click Load unpacked and choose the extension/ folder, then paste the key in the options page and save. The key stays in your browser's local extension storage; refresh linkedin.com/feed afterwards.

Yes. The README ships an offline harness: run cd extension and python3 -m http.server 8765, then open http://localhost:8765/fixtures/feed.html. That fixture path needs no key and costs nothing.

At the time of writing Jev charged $0.042 per million input tokens with free output; the README says a full 30-post evaluation run used about 37k input tokens. It tells you to check the current TypeSafe pricing yourself.

Accuracy is unproven — the 30 included examples are synthetic sanity checks, and the README says to judge the filter against your own labeled posts. LinkedIn prohibits extensions that modify its appearance, the build is not distributed through any store, and it only reads and blurs; it never likes, comments, messages, or posts.

Edit the rubric questions in extension/src/shared/rubric.js and the thresholds in decide.js, bumping the version constants when you do. To measure it, label some posts yourself and run them through the offline harness in eval/harness/; the README warns never to ship a build with a shared key embedded.