Guide
How to check a crypto backtest: ten questions
Ask them of any backtest: a bot’s, a newsletter’s, or ours. Each answer below is Fygga’s own, including the mistakes.
Why these: in 30 public issue threads from two open-source bot projects (Freqtrade and Hummingbot) that we coded in September 2026, 23 were about order-state reliability, accounting, or the integrity of backtests and paper trading. That is a small sample, not a survey, but the pattern is the point.
01
Does every number name its window?
A return without a start date, an end date and a length is undefined, not just weak.
On Fygga: Every figure carries its window; the headline run is 1 Jan 2022 – 24 Jun 2026. Results
02
Is it a total return or an annual one?
The same figure reads several times larger if a total is mistaken for a yearly rate.
On Fygga: +2.8% in total over 1 Jan 2022 – 24 Jun 2026 is 0.6% CAGR. Our own research notes once recorded it as “a year”; the site now prints total and CAGR separately.
03
Are costs charged on both legs, on every rebalance?
A hedged trade pays fees and slippage on the spot leg and the futures leg, each time it moves.
On Fygga: Both legs, every rebalance. Methodology
04
Is there a cost sweep, or one flattering assumption?
A result that only works at the cheapest fee tier is a result about fees.
On Fygga: Four fee tiers, zero cost included, side by side. Sub-periods
05
How often does it rebalance?
Trading more often pays costs more often. Cadence can flip the sign of a result on the same data.
On Fygga: Same data, same window: +2.8% rebalancing weekly, -19.1% rebalancing daily.
06
Does it hold up in sub-periods, or only on average?
A positive four-year average can hide a recent period that loses money.
On Fygga: The full window is positive at the stated costs; at the same costs, the first half of 2026 is negative at every cadence. Verdict
07
Was margin tested?
A hedged pair can still be closed out: the futures leg is margined on its own.
On Fygga: At 5× from a 2022 start, none of 16 cost and rebalancing combinations survived our margin test.
08
What happens to missing or partial data?
A day with one funding settlement out of three is not a complete day. Summing it as one changes the result quietly.
On Fygga: Complete days only. 4 Sep 2026 was left out for that reason.
09
Is the forward record ever back-filled, and does stale data say so?
A paper record filled in after the fact is a second backtest, not a record.
On Fygga: No data, no point; stale after 48 hours, and the page says so. A silent error once froze our record for 22 days. Paper record
10
Was the pass/fail rule written before the test, and are the rejects shown?
A rule chosen after seeing the result can always be made to pass. A page of winners is a selection.
On Fygga: The 2026 rule was written first and the strategy failed it; the candidate rejected earlier is published too. Verdict · Compare
Before any backtest: what can the tool touch?
“Non-custodial” covers four separate questions. Fygga’s answers:
| Where are your funds? | Fygga holds none. |
| Where is your exchange key? | Never asked for. Fygga never connects to your exchange. |
| Can it place trades? | No. Fygga places no orders; it is paper only. |
| Can it withdraw? | No. There is nothing connected to withdraw from. |
How other products answer the same four: where crypto-bot API keys and funds live.
Educational only — not financial advice.