Anyone can build a trading bot. Almost no one tells you the truth about whether it actually works. Tessen's engine grades every agent out-of-sample, net of every fee, tested for robustness — a real verdict, not a backtest you can talk yourself into believing.
An agent doesn't go live or get listed for sale unless it clears all five. No exceptions, not even for us — we graded our own strategy first and it got a D.
Chronological 60/40 split, after fees, funding, and slippage — not a curve-fit vanity backtest.
Gross edge ≥ 1.5× round-trip cost. No strategies that only "work" by scalping inside the spread.
Positive across ≥60% of assets, majority of regimes, majority of yearly windows — not one lucky run.
Max drawdown ≤ 35% at responsibly-sized position risk — Monte-Carlo tested, not eyeballed.
Test-set expectancy ≥ 50% of train-set expectancy. If it falls apart out of sample, it fails.
Plain English: "buy BTC when RSI drops below 28 on the 15-minute chart, take profit at 3%, stop at 1.5%." No code required.
One structured contract every builder — conversational, visual, or code — compiles to. The same spec runs in backtest, paper, and live.
Out-of-sample, net-of-fees, fill-realistic, robustness-tested. A real letter grade — computed by the engine, never self-reported.
Edit a rule, watch the Grade update live. Ship to paper trading, connect your own exchange, or walk away knowing before risking real capital.
This is the Studio — a mock walkthrough, not live data. Every real agent goes through this exact screen before it ever touches a chart.
Non-custodial by design: agents trade on your connected exchange with a trade-only key. Keys with withdrawal permission are rejected outright — Tessen can never move your funds, and neither can anyone who compromises us. Three modes: dry-run (simulated), confirm (approve every trade in Telegram), or fully automatic live.
6 exchanges connected today: Binance, Bybit, OKX, Bitget, Coinbase, Hyperliquid.