by Zaious · Agent Tool · ★ 28
🇹🇼 中文(本頁)|🇬🇧 English Jev Capability Atlas 獨立、非官方、非 TypeSafe 贊助的社群專案。 用真實 API 呼叫的收據,畫出 Jev(TypeSafe 的 System One 模型)「校準決策」這個宣稱在哪裡站得住、在哪裡站不住的邊界地圖——給人看怎麼用,給 agent 帶進專案看哪裡能試著換上去,也給任何做過真實驗的人一個把結果貢獻進來的地方。 不是排行榜(市面上已經有 jev-benchmarks、thaiexam-jev-charts 在做這件事,做得很紮實,我們引用它們、不重做)。這裡要回答的是更根本的問題:什麼情況下它強,什麼情況下它弱,為什麼。 30 秒版 Jev 很快、很便宜,只能做「選一個選項/打個分/回答是非」這種窄判斷,不會寫文字解釋自己在想什麼。因為答案空間是你自己先定義好的,它結構上不可能吐出選項清單以外的東西——這跟自由生成文字的模型偶爾格式跑掉、甚至生出一個你沒列的分類,是不同等級的保證,不是機率低,是型別上不可能(但這只保證答案落在清單裡,不保證選到的那個是對的,細節見下面「不是瞎猜」那節)。...
| Stars | 28 |
| Forks | 6 |
| Language | Python |
| Category | Agent Tool |
| License | MIT |
| Quality Score | 58.6641152026131/100 |
| Open Issues | 2 |
| Last Updated | 2026-09-24 |
| Created | 2026-09-18 |
| Platforms | python |
| Est. Tokens | ~14k |
These tools work well together with jev-capability-atlas for enhanced workflows:
Explore other popular agent tool tools:
jev-capability-atlas is Independent, evidence-based map of when TypeSafe's Jev actually holds up vs. breaks down — real API-call receipts, not a leaderboard. 中文為主的雙語 repo。. It is categorized as a Agent Tool with 28 GitHub stars.
jev-capability-atlas is primarily written in Python. It covers topics such as ai-agents, benchmark, calibration.
You can find installation instructions and usage details in the jev-capability-atlas GitHub repository at github.com/Zaious/jev-capability-atlas. The project has 28 stars and 6 forks, indicating an active community.
jev-capability-atlas is released under the MIT license, making it free to use and modify according to the license terms.