Bitra
Autonomous trading research agents whose final verdicts, event guards and reviewer cases are typed Jev decisions.
Autonomous trading research agents whose final verdicts, event guards and reviewer cases are typed Jev decisions.
Jev is the typed judge. Headlines are read as closed questions: which kind of event they describe, whether they concern the instrument, and whether they speak to the setup. Each cycle the two reviewer cases, already checked against the figures they cite, are weighed in one Pick over support, oppose or abstain. That verdict is the one the system counts; sizing, risk limits, the exit and the order stay in deterministic code.
Weighing only the two checked cases, which verdict on the proposal do they support?
One trade proposal with two advocates' cases, each claim already checked against a cited figure. The repository defines the Pick over support, oppose and abstain, and code records an abstain as an answer rather than an error.
The engine fetches candles, the book, funding and open interest into a snapshot; agents read the snapshot and never fetch.
The event guard asks Jev which kind of event a headline describes and whether it concerns the instrument.
A structure lead proposes a trade; two advocates build a case for and against it from the same figures.
The system verifies each cited claim against its figure and removes ones that fail before the judge sees them.
Jev weighs the surviving cases and answers support, oppose or abstain with a probability.
Sizing, the exit, the order and every risk limit are applied by code after the verdict.
It separates a reasoning model that writes the cases from a decision model that chooses between them, and it refuses moving model aliases so a fitted threshold cannot drift. A judge failure is treated as an absent answer, never as a no.