Overview & result
SuperTurbo shows Hawk or dove?, built on Jev: Fed press conferences scored word by word, Jev against chat models.
From the original post
Hawk or dove? by @superturbo: Fed press conferences scored word by word, Jev against chat models.
How Jev fits in the loop
- The page plays a Fed press conference with the official transcript as captions and judges only the Chair.
- Jev scores each word against four stances: Hawkish, Partly hawkish, Partly dovish and Dovish.
- A chat model scores each caption line against the same stances for comparison.
- Answers are saved, so each past conference is paid for once and replays free.
How to reproduce
- Review the source documentation and repository at https://github.com/TurboGuo/jev-fedspeech.
- Configure environment with TypeSafe Jev / System One API keys.
- Run test prompts against the decision endpoints.
Why this build matters
Shows Jev in practical use for trading and financial markets.
Limitations
- Not affiliated with TypeSafe, the chat-model providers or the Federal Reserve; the author states it is not financial advice.
- The speed figures (about 150 ms per word for Jev, a few seconds per line for the chat model) are the author's.
- The recorded demo replays saved answers and makes no API call.