Question 1

Does this send my timeline to a model vendor?

Accepted Answer

Extracted post text and your interests go to your Cloudflare Worker, and on a cache miss to TypeSafe. Author handles are used locally for the allowlist and are not separately sent for scoring, but handles or personal details contained inside post text can still be transmitted. The KV cache stores scores rather than post text, and the cache key covers the rubric version, the text and your interests.

Question 2

Will it hide a post I actually wanted to read?

Accepted Answer

Every collapsed post carries a Show anyway button for the current page session, and a dimmed post can be peeked at on hover. Collapse needs both a net score at or below -0.90 and minimum confidence across the used dimensions of at least 0.30, so a low-confidence result can be dimmed but never collapsed. You can also allowlist authors, which skips scoring and filtering for their new posts.

Question 3

What are the five sliders actually measuring?

Accepted Answer

Firsthand experience, self-promotion, engagement bait, technical depth and relevance, each scored 0 to 1 and then combined into a weighted net where zero-weight dimensions are ignored. They are content judgments, not fact-checks and not an AI-slop detector, and the rubric itself lives in worker/src/questions.ts. The debug strip shows the per-dimension scores, the net and the proposed action, with the raw score and confidence JSON on hover.

Question 4

Is running this inside my own browser allowed by X?

Accepted Answer

The README calls it a personal-use prototype and says to review X's current terms and automation policies before using or distributing it. Running inside your own browser does not establish permission from X. It also leaves your feed order alone: posts are dimmed or collapsed in place, with no summaries and no reordering.

Question 5

What are the batching and caching limits?

Accepted Answer

The extension sends up to 12 posts per batch and up to 640 characters per post. The Worker caps text at 1,000 characters, scores up to four posts concurrently, and caches scores in KV for 30 days, with KV values holding scores rather than post text.

Question 6

What is the exact decision rule?

Accepted Answer

net is the sum of (slider / 3) times each dimension score, with sliders from -3 to +3 and dimension scores from 0 to 1. net at or below -0.30 dims the post and at or below -0.90 collapses it, unless the minimum confidence across used dimensions is under 0.30. Zero-weight dimensions are ignored.

Question 7

What does it not process?

Accepted Answer

It runs only on the home timeline and lists, so DMs, notifications, profiles and search pages are skipped. Protected-account posts are not detected, only text is analyzed, and long text is truncated, so images, video, replies and full thread context are not judged.

Jev X

About this project

Highlights

Questions about this entry

More Jev resources like this

From the Jev guides

From the Jev community