按症状诊断 Jev，而不是靠随机猜测。
把 FAQ 当成诊断中心来用：先从失败现象出发，缩小可能原因，再决定是否改 setup、模型或工作流。
- 8 个症状条目
- 先从失败现象出发，而不是靠随机猜测
- 在改 setup、模型或工作流前先缩小可能原因
API calls fail auth.
I get 401 / unauthorized
Check key presence, path (TypeSafe vs Gateway), and Bearer formatting.
Thinking the skill is mandatory.
Is the agent skill required?
Optional accelerator for coding agents.
Automating on every answer.
Can Jev be wrong?
Yes — use confidence + human paths.
Repeating marketing numbers as our benchmarks.
What about 193× / 444× claims?
Those are TypeSafe-published workflow-eval claims — attribute, do not re-run here.
People expect chat replies from Jev.
Is Jev an LLM?
No — it returns structured decisions, not chat.
Confusion about Noul vs Choice confidence.
What is Noul?
Noul is P(yes) on 0–1; see primitives docs.
Treating SemIf scores as hosted Jev accuracy.
OpenJev / SemIf vs Jev?
SemIf is an interface pattern on open models — not TypeSafe weights.
Blocked before first experiment.
Does the Playground need signup?
Use console playground and/or community mirrors; re-check live access rules.