A backtest shows what a strategy did under specified assumptions; it does not show what live trading will produce. Before trusting one, audit its slippage, pricing inputs, subscription costs, drawdown, and any points where a human had to intervene.
In 1970, Apollo 13’s crew faced a problem that had not appeared in the original flight plan: an oxygen tank failure left the mission operating under conditions its earlier assumptions did not cover. NASA engineers had to work with the spacecraft’s actual constraints, including a carbon dioxide buildup and equipment that was never designed for that exact combination of problems.
The mission’s outcome was still uncertain while engineers developed a workaround. The crew returned safely, but the result depended on adapting to live conditions rather than trusting the original plan. The incident is documented in NASA’s Apollo 13 mission records and in Jim Lovell and Jeffrey Kluger’s Lost Moon.
A trading backtest creates a similar risk when its clean historical result depends on inputs that will change at execution. The strategy may look stable until the market supplies a price gap, a wider spread, a delayed fill, or a decision point the test did not model.
Slippage and pricing inputs can change the result
Start with the entry and exit prices. Ask whether the backtest assumes that every order fills at the last quoted price, the candle close, the midpoint, or a price that was actually available to a trader.
That distinction matters most in volatile markets and for larger orders. A strategy that buys at the displayed price in a backtest may receive a worse fill in live trading. If the expected gain per trade is small, a modest difference between the assumed price and the executed price can remove much of the edge.
Check whether the test includes:
- Bid and ask spreads
- Partial fills
- Delayed execution
- Gaps between bars
- Market impact
- Trading fees
- Different liquidity conditions
Also check the data itself. Were prices adjusted for stock splits and dividends? Did the test use information that would only have been known later? Were delisted assets excluded? A backtest can produce a strong result while quietly removing the situations that would have weakened it.
Composer’s September 2026 documentation update is useful because it puts these assumptions in view. Treat the documentation as an audit list, then ask which assumptions match the account, market, and order size you plan to use.
Drawdown matters more than the smooth line
A high ending balance can hide a drawdown that would have forced a real trader to stop. Review the largest peak-to-trough loss, how long recovery took, the number of consecutive losses, and whether the worst period occurred near the beginning or end of the test.
A strategy with a tolerable average return may still be unusable if its drawdown exceeds your predefined loss limit. For a smaller account, position sizing can make the same percentage decline feel materially different. For a larger account, liquidity and execution may become the bigger constraint.
Do not inspect only the total return. Compare the result across different market conditions and time periods. A strategy that works during one strong trend may behave differently during a range-bound or sharply declining market.
This is also where an approval gate can add discipline. An AI assistant can generate and queue a signal, while the trader checks the current exposure, planned loss, and market conditions before approving the order. That approval does not improve a backtest by itself. It makes the gap between a historical model and a live decision visible.
Subscription costs and intervention markers belong in the test
Include the recurring cost of the tools required to run the strategy. Subscription fees reduce net performance, especially when the account is small or the strategy trades infrequently. If a backtest reports gross results, calculate the net result after platform fees, data costs, commissions, and other known expenses.
Then look for intervention markers. Did the strategy require a manual pause, parameter change, skipped trade, or discretionary override? Mark those events instead of treating them as invisible exceptions.
A clean equity curve can depend on decisions that were made outside the stated rules. If the operator removed certain trades after seeing what happened, the test no longer represents a rules-based process. That does not make the analysis useless. It means the human judgment needs to be documented and evaluated separately.
The same discipline applies when reviewing a live signal. Before approving it, compare the current price with the price assumed in the test, confirm the position size, and check whether the trade would breach a daily or portfolio loss limit. Backtest Assumptions: Why Evan Rejected Monday’s Live Signal explores that decision at the order level.
Turn the audit into a repeatable approval step
Create a short record for every strategy:
- Assumed entry and exit prices
- Slippage and fee assumptions
- Data source and adjusted-price treatment
- Maximum drawdown and recovery period
- Subscription and execution costs
- Manual interventions and skipped trades
- Position-sizing and loss-limit rules
Update the record when the market, account size, or trading tool changes. A backtest is evidence about a model under defined conditions. It becomes more useful when those conditions are visible and compared with the conditions in front of you.
Apollo 13’s engineers had to stop relying on the original assumptions once the spacecraft’s situation changed. A trader reviewing a queued order faces the same basic discipline: inspect the live constraints before approving the plan. Educational content, not financial advice.
Comments
No comments yet.