A bot has evidence of edge only after realistic fees, spread, slippage, latency and failed-order behavior are included. Paper trading should validate execution assumptions before capital is exposed.
Three-step decision framework
1. Data integrity
Check timestamps, missing intervals, survivorship and look-ahead bias.
2. Cost model
Model maker/taker behavior, spread, slippage, funding and rejected orders.
3. Staged validation
Move from replay to out-of-sample, paper and tightly limited live exposure.
Validation path
- DCA Backtest — See how assumptions change historical outcomes.
- Slippage Calculator — Estimate execution loss beyond headline fees.
- Funding Arbitrage Calculator — Model carry return and margin assumptions.
- Risk / Reward — Define acceptable loss before automation.
- Research Library — Review methods, limitations and market evidence.
Frequently asked questions
Why do backtests outperform live bots?
Backtests often underestimate latency, slippage, queue position, partial fills, outages and changing market structure.
What should paper trading prove?
It should test signal timing, order lifecycle, data gaps, cooldowns, reconciliation and monitoring—not just theoretical P&L.
How much history is enough for a bot?
There is no universal number. The sample must cover enough trades and multiple market regimes for the strategy horizon.
This page provides research and decision frameworks, not investment, legal or tax advice.
