In those five minutes, treat the AI signal as a proposal, then compare its assumptions with current price, total account risk, and the conditions that would invalidate the trade. Approve it only when the planned entry, position size, and exit still fit; reduce it when the setup remains valid but the exposure is too high; reject it when the evidence or risk limit has changed.
In 1983, Lieutenant Colonel Stanislav Petrov was on duty at Serpukhov-15, a Soviet early-warning command center near Moscow, when the system reported an incoming United States missile. It later indicated additional launches. The warning demanded attention, but Petrov did not treat the machine’s output as sufficient proof.
He assessed the signal against its context. A small number of reported missiles did not fit the pattern he expected from a first strike, and ground radar had not confirmed the launches. He reported the warning as a false alarm. The system had misread sunlight reflected from high-altitude clouds.
David E. Hoffman documents the incident in The Dead Hand. Petrov did not know the outcome when he made the call. He had a machine-generated warning, incomplete confirmation, and consequences that made blind acceptance dangerous.
A trade signal carries vastly smaller stakes, but the decision has the same structure: automation detects a pattern; a human must decide whether the evidence and context justify action.
Separate the signal from the decision
An AI signal can identify a setup without knowing everything that matters to you. It may detect momentum, a breakout, or a statistical relationship. Your account still has its own limits: available capital, open positions, recent drawdown, correlated exposure, and the amount you can lose without changing your behavior on the next trade.
That distinction matters before market open, when price can move quickly and urgency can disguise missing information.
Start by translating the signal into a falsifiable trade plan:
- What price makes the entry valid?
- What price proves the idea wrong?
- How much would the position lose at that level?
- Does the trade add exposure already present elsewhere?
- Has pre-market movement changed the entry enough to alter the original calculation?
If those questions have no clear answers, approval would mean accepting uncertainty you have not measured. The countdown to the open does not repair an incomplete plan.
Use three decisions, not one
A binary choice between “take the trade” and “miss the trade” creates avoidable pressure. The useful decision set has three options: approve, reduce, or reject.
Approve when the current price still supports the original reasoning, the invalidation point remains clear, and the full position stays within your risk limit. Approval should preserve the plan as written. If you need to move the exit farther away to make the trade feel safer, the planned loss has changed.
Reduce the size when the setup remains intact but full exposure no longer fits. Perhaps the opening gap widened the distance to invalidation. Perhaps another position already responds to the same market factor. Smaller size can keep the loss at the planned amount without pretending the added risk disappeared.
Reject when the market has removed the edge, confirmation is absent, or the trade would break an account-level rule. Rejection is also appropriate when you cannot explain why the signal remains valid. The reasoning matters because “the AI found it” does not define when the trade is wrong.
A zero position is a valid position size. Marcus used that choice during a prop evaluation because preserving the risk limit mattered more than participating in one setup.
Make the five-minute review mechanical
Five minutes is enough for a short review if the limits were set earlier. It is not enough to invent risk rules while watching the opening price.
At five minutes before the open, record the proposed entry, invalidation level, and planned loss. Check existing positions for overlapping exposure. Then compare the live price with the assumptions behind the signal.
At three minutes, calculate the position size from the distance between entry and invalidation. If the loss exceeds your limit, reduce the size or reject the trade. Do not solve the problem by moving the invalidation point without new evidence.
At one minute, state the decision and its reason in one sentence. For example: “Reduced because the wider entry-to-exit distance would exceed the planned loss at full size.” That sentence belongs in the trading journal beside the signal and eventual outcome.
This record prevents outcome bias later. A rejected trade that rises can still be a correct rejection. An approved trade that loses can still reflect a sound process. Judge the decision by the information available before execution, then use repeated results to evaluate the signal source.
For a concrete example of protecting a fixed loss limit, see how Eli handled an AI trade rejection.
Keep the human gate consequential
An approval button has little value when approval is automatic in practice. The gate works only when rejection leaves the order unexecuted and reduction changes the exposure before execution.
Petrov’s decision at Serpukhov-15 mattered because he assessed the warning instead of forwarding it as unquestioned truth. Your review should do the same work at a smaller scale. Compare the signal with independent evidence, account constraints, and a written invalidation point.
Before the next market open, write three lines at the top of your journal: full-size conditions, reduced-size conditions, and rejection conditions. When the signal arrives, choose the line it actually meets.
Educational content, not financial advice.
Comments
No comments yet.