用 Jev 做时间线分析器：每条帖子 12 个结构化判断

Defileo🔮 @defileo原推发布于 2026.09.23

在 X 上打开原推

Nothing is impossible with Jev now, I just proved it myself. I built a Jev X Feed Analyser, it reads my timeline while I scroll and judges every post on the screen Here is how it actually works: > every post gets the same 12 typed questions > not a summary, not a vibe check, a schema > hook_archetype, topic, reach_band, is_shill, reads_like_ai, claim_needs_source, regret_if_quoted, worth_reading, worth_reply, bookmark_worthy, opens_with_number, mute_author > each answer is a bool, a score or a choice from a fixed list > each answer carries its own confidence, 0 to 1 > 203ms average per answer, so a full post is judged before I finish reading it The output is not prose, it is a stream of typed JSON, that is the whole point: {"is_shill": 0.36, "confidence": 0.44, "ms": 91} {"mute_author": 0.80, "confidence": 0.46, "ms": 276} A model that writes paragraphs cannot be sorted, filtered or trusted twice, a model that emits typed fields with confidence can be turned into a rule, and one scroll proved it: > 30 posts, 1,318 judgments, keep 10, reply 4, skip 10, mute 3, 52% of my feed was worth the scroll, the rest I never have to see again Then I ran the same 12 questions across a batch corpus of 100,000 viral posts, to have a baseline for what a good post even looks like > 16.5M tokens in > $0.042 per million > total cost, $0.693 > output tokens are free 7 hook types ranked by 1M+ reach, superlative claims are most common, and convert poorly, receipts/proof drops are used half as much, yet hit the top 1% more often. Your feed is not an algorithm problem, it is a labeling problem, Jev just made labels cheap enough to run on everything you read

这条是引用着另一条写的：x.com/1663262095868084228/status/210214035969030…