Jev runs through years of bookmarks

Just a video post. The whole public artifact is a caption saying that years of Twitter bookmark hoarding is light work for Jev to run through, plus a signup link for the author's separate transcription product. The source documents no repository, no prompt, and no way to reproduce the run.

The post does not say. No bookmark count, latency, cost, or per-item decision type appears anywhere in the source. The caption implies a good fit for bulk classification, but nothing there tells you whether it was one batched pass or one call per bookmark.

Undocumented. There is no criteria list, no label set, and no state format in the source, so you cannot tell whether it was a Choice over keep and discard, a Score, or a Noul. That prompt shape is the part that decides whether the approach works, and it is exactly what is missing.

You could not, from this. No precision, recall, or spot-check figures are published, and the author calls the workload light work rather than a measured result. The only replies are a joke about bookmarking the post itself and unrelated signup promotion.

The follow-up points to getmargin.vercel.app, where the author asks you to sign up so he can email you when it is live. The bookmark run is the teaser for that product, not a released tool.

The post is dated 20 September 2026 at 2:05 PM and shows 185.1K views with 57 replies, 110 reposts, 3.1K likes and 4.1K bookmarks. Those are the only numbers attached to it.

One reply jokes that the commenter will bookmark the post and surely get back to it, just like their other bookmarks, which is the whole gag. Nothing in the thread supplies a bookmark count, a prompt, or a result.