JEV EXPERIMENTS

A referee, not a filter bubble.

Chrome extension / v1.1.0

Feed
Referee.

Less bait. Your call.

Jev flags ragebait, promo, AI Slop, and engagement farming while you scroll X. Keep the labels, or collapse the noise.

Desktop Chrome / Chromium. Bring your own TypeSafe key.
Opens GitHub. Install unpacked, not from the Web Store.

In the wild / X00:38 · Real footage
Jev caught this site’s own developer lacking.
Read the demo walkthrough
  1. Enable Feed Referee in the popup.
  2. Open a Home badge: Promo, 90%.
  3. Check bookmarked posts: AI Slop, 77%, and Ragebait, 80%.
  4. Open a profile post with two labels: Promo, 89%, and Engagement-farming, 82%.
  5. Turn on optional collapsing, then click Show tweet to restore a hidden post.

Four calls. Not one verdict.

Each label is judged independently. Badges appear at 70% or above, and one post can get more than one.

01

Ragebait

Provocation aimed at anger or hostile engagement. Not just an opinion you disagree with.

02

Promo

Pushing a product, service, project, or signup. Promotion isn't necessarily a bad thing.

03

AI Slop

AI-like writing patterns and canned filler. A style estimate, not proof of AI authorship.

04

Engagement-farming

Fishing for replies, likes, reposts, or follows rather than sharing something of substance.

[ On your terms ]

A second opinion.
Not a disappearing act.

Rohan’s personal-assistant invites post, with Feed Referee estimating Promo at 96% and Engagement-farming at 89%.

01 / In the name row

A badge, not a lecture.

Compact labels sit alongside the author. Click for probabilities and what each label means. No feed reordering, no auto-likes, no auto-replies.

Popup settings with optional collapsing enabled for all four labels and the minimum probability set to 85%.

02 / You make the call

Hide less. Or hide nothing.

Collapsing is off by default. Choose labels and a threshold in the popup. Changes autosave, and Show tweet always brings a collapsed post back.

Screenshots from the real demo. Click to enlarge.

[ Get it running ]

Your browser.
Your key.

No Jev Experiments account. The extension talks directly to TypeSafe, and API usage is billed to your TypeSafe account.

Get a TypeSafe key ↗
  1. 01

    Download & unzip

    Open the GitHub download instructions ↗. Get the Feed Referee ZIP from Releases and extract it. Keep the extracted folder somewhere permanent.

  2. 02

    Load the extension

    In desktop Chrome, open chrome://extensions. Enable Developer mode, choose Load unpacked, and select the extracted feed-referee folder containing manifest.json.

  3. 03

    Add your key. Then switch it on.

    Pin Feed Referee. Paste your own TypeSafe key, review the data-sharing notice, and set a daily request cap. It starts paused. Enable it, then open or reload X.

[ The fine print, up front ]

Know the limits.

Read privacy & limits →
Where does it work?

On supported, visible posts across X: Home, profiles, post pages and replies, search, lists, and bookmarks. It checks the main column in active tabs. Individual post pages show labels without collapsing the post or its replies. X layout changes can require an extension update.

Does it read my DMs or drafts?

No. DMs, composers, drafts, account/settings screens, and dialog content are excluded. It judges the text X has loaded, plus quoted text for context. No images, videos, or linked pages are analyzed. Expanding a truncated post can update its judgment.

Where does my data go?

When enabled, post text goes directly from the extension to TypeSafe using your own API key. This can include protected posts you can view. No requests go through Jev Experiments servers, and the extension has no analytics or telemetry. TypeSafe's own terms and privacy policy apply.

What does it cost?

The extension download is free. Jev requests are billed to your TypeSafe account. New installs have a 2,000-request daily cap that you can lower in the popup. Attempts, including failures and retries, count across tabs and reset at midnight UTC. This is a request cap, not a dollar spending limit.

Can I trust every label?

No. These are model probability estimates, not verified facts. Context, irony, and language can affect the result. AI Slop does not prove who or what wrote a post. Nothing is hidden by default, and every collapsed post has a Show tweet button.

Is this in the Chrome Web Store?

Not yet. Download it from GitHub and use Chrome's Load unpacked option. It requires desktop Chrome 120+ or a compatible Chromium browser, plus your own TypeSafe key. Updates are manual: replace the unpacked files, then click Reload on the extension card.

Give your feed a second opinion.

Source available. Your own key. Always your call.