What jarrodwatts/jev-trader documents
jev-trader describes using the Kuru MON-USDC order book on Monad, selecting a side and maintaining post-only limit orders. Its README distinguishes the configured decision model from a mock momentum policy and distinguishes dry-run mode from signed execution.
The documented default is a mock policy, and the linked public deployment is described as a dry-run mock deployment. Seeing its chart move is therefore not evidence that a live Jev call made the decision or that a trade filled on-chain. Verify the model identity and execution mode in the actual records.
Keep the model judgment separate from the controller
market snapshot + bounded question -> model judgment
-> deterministic eligibility and risk checks
-> simulated or real order submission
-> execution records and position reconciliationThe model can judge the supplied state. Code owns numeric arithmetic, position limits, order validity and whether execution is enabled. The Jev 1.13 guidance specifically recommends keeping precise arithmetic in code.
Log the observation time, model version, answer, controller decision, order identifier and resulting fill or cancellation. If a source, inference request or execution step fails, its failure must remain visible instead of turning into a fabricated market decision.
A realistic research comparison
Use the same market observations, decision times, execution rules and constraints for the compared policies. Include fees, latency, cancellations, missed fills and unavailable requests. A favorable decision before those effects is not the same as a completed trade outcome.
Keep strategy tuning separate from out-of-sample evaluation. Avoid using information that was not available at the original decision time. Report the test window and market conditions; one short run cannot establish behavior in other conditions.
Mock, replay, paper and live modes
| Mode | What you can inspect | What remains unproven |
|---|---|---|
| Mock policy | Interface and controller plumbing. | Actual model decisions. |
| Historical replay | Decisions under recorded observations. | Live data and realistic execution unless explicitly modeled. |
| Paper or dry run | Live observation and simulated actions. | Actual fills, transaction costs and realized outcomes. |
| Live execution | Actual order and fill records. | Future profitability or reliability across regimes. |
How to read a trading project's claims
Ask which decisions came from Jev, which model version answered, which results were simulated, and whether the execution and accounting records reconcile. Inspect limitations as closely as headline performance numbers.
This is an engineering reading guide for a community experiment. JevHunt has not operated the bot, reproduced its trading results or established a return expectation. The decision-model evaluation guide provides a general scorecard, but model accuracy and trading outcomes are different measurements.