How to Choose a Backtesting Platform Before You Trust a Curve
A backtest is a research instrument. Pick the engine that can lie to you in ways you can audit — lookahead, fills, and costs — then read the mistakes guide.
Published: Aug 24, 2026 · Updated Aug 24, 2026
Author: Traderfolio
Editorial standards. Guides are written from published catalog data and vendor documentation — not paid placement copy. Sponsored listings are labeled separately. See listing guidelines.
Backtest marketing is a Sharpe on a slide. The product is an event loop plus a fill model. If you cannot say how the bar was closed before the signal was known, you do not have research — you have a trailer.
Use the algo hub and the backtesting collection. Pair this with backtesting mistakes. The three-engine compare is QuantConnect vs Freqtrade vs LEAN.
1. Name the market and the hold time
Write:
- Asset class and venue
- Bar vs tick
- Intraday vs daily
- Whether you need live paper after the test
- Language you will still be able to read in six months
A crypto bot on 1-minute candles and a daily equity factor study are different engines.
2. Audit the three lies first
Every platform can lie. Prefer the one that makes the lie visible.
- Lookahead — is the signal using the close that was not knowable?
- Fills — next bar open, mid, or a fantasy last?
- Costs — spread, fees, borrow, funding?
If the docs hand-wave “realistic fills,” assume they are not. The mistakes article lists the rest.
3. Data rights are part of the platform
A free engine on a forbidden CSV is still a license problem. See how to choose a market data feed. Confirm whether the platform’s included data may be stored and whether it matches the live venue you will trade.
4. Live is a second purchase
Research notebooks, hosted clouds, and self-hosted bots do not share an operations story.
- Who runs the process overnight
- How you promote a parameter change
- Whether the live fill path is the same object as the backtest
If promotion is “export and hope,” you will fork reality.
5. Compare two engines on the same spec
Write a one-page spec: universe, signal, costs. Implement it twice from the collection. If the curves only match after you “fix” one engine, you do not have a result — you have a negotiation.
Discuss infrastructure in Algo & Quant after you have a spec. The room will not debug a curve you cannot reproduce.
FAQ
Can I trust a hosted Sharpe?
Only as a number the host computed on their data and their fill model. Reproduce it or ignore it.
Do I need code?
If you cannot state the fill rule, a click-together builder will hide the lie. Code is optional; auditability is not.
Where is the aisle?
Algo trading. Featured is editorial, not edge.