Scrutinizing Jev's trading-bot latency and cost claims
The author lists Jev's claimed 70-500ms decisions and 40-400x cost savings, then demands eval tables, latency histograms, decision logs and backtest comparisons.
jev: 70-500ms decisions. $0.042/m input tokens. 20-200x faster. 40-400x cheaper. 67.8% agreement with frontier consensus at 76x lower cost. live monad bot at 81ms. $10-25/month for 24/7 block loop. receipts missing: • exact typeSafe four-workflow eval table (model versions, sample size, agreement definition, raw scores) • latency histogram from the jarrod watts bot (p50/p95/p99, not a single 81ms claim) • the 1000+ stars repo with the actual decision logs and fill data • gas vs spread math that shows the 428 mon/hour loss (inputs, assumptions) • any public calibration curve, brier score, or ece on real order-book data • side-by-side: same strategy with jev vs rules vs llm, with sharpe, drawdown, adverse selection numbers decision engine ≠ trading system. latency and cost claims are directional until the tables and the live run logs land. post them.
This post quotes another post: “Building an HFT trading bot with Jev: one buy/sell decision per block”