A trade can follow every entry, sizing, and execution rule and still leave you holding a position you never wanted. Perfect execution only proves that the system followed its instructions; it does not prove that the instructions matched your risk limits or intent.
In September 1999, NASA’s Mars Climate Orbiter approached Mars after a journey of roughly nine months. Engineers expected the spacecraft to enter orbit and study the planet’s atmosphere. Instead, communication stopped.
The commands had been processed. The calculations had been performed. The spacecraft had responded. Yet one part of the navigation software produced data in pound-force seconds while another expected newton seconds. NASA’s investigation concluded that the unit mismatch contributed to a trajectory that brought the spacecraft too close to Mars. The mission was lost.
NASA’s Mars Climate Orbiter Mishap Investigation Board Phase I Report documents the failure in direct terms. The machinery did what the system told it to do. The larger problem sat upstream: the system’s components did not agree on what the numbers meant.
Correct execution can preserve the wrong assumption
Trading systems can fail in the same shape.
Suppose an AI trading assistant identifies a breakout, calculates an order, and queues a long position. The setup meets its technical criteria. The order size follows its configured rule. The entry occurs at the specified price.
Then you review the total exposure.
You already hold two positions that move with the same market. The new order would concentrate more of the account in one direction. Perhaps the proposed stop is valid for the chart but would create a loss larger than you want to accept. Perhaps a scheduled event changes the conditions under which you are willing to hold the trade.
None of those facts makes the signal’s execution defective. They make the resulting position unacceptable to the account holder.
That distinction matters because traders often audit the wrong layer. They ask whether the model followed its rules, then treat a “yes” as evidence that the trade belongs in the account. The Mars Climate Orbiter followed instructions too. The failure came from a mismatch between those instructions and the system’s actual requirements.
A valid setup can still violate the account
A trade has at least two separate tests.
The first asks whether the market setup is valid. Did price meet the entry criteria? Is the stop placed according to the strategy? Does the expected reward justify the defined risk under the system’s assumptions?
The second asks whether the trade is acceptable now, inside this account. How much capital is already exposed? How closely does the new position move with current holdings? What happens if all related positions reach their stops during the same move? Would taking this trade violate a session loss limit or create a position you would resent owning five minutes later?
Passing the first test does not guarantee passing the second.
This is why position sizing needs an account-level view. A 1% risk allocation may look disciplined in isolation. Three correlated positions, each carrying 1% of account risk, can behave more like one concentrated bet when the shared market driver reverses. The arithmetic stayed clean. The assumption of independence did not.
The same issue appears when a system re-enters after a stop. Each trade may satisfy the strategy, but the combined sequence can exceed the loss you were prepared to accept for that session. Daniel’s Bot Re-entered. One More Stop Could End His Session. examines that cumulative risk.
Approval is a risk control, not a ceremonial click
An approval gate creates a deliberate point between signal generation and order execution. Its value depends on what happens at that point.
Blind approval adds little. Useful review compares the proposed trade with constraints the signal may not fully represent:
- Check the loss at the stop in currency and as a percentage of the account.
- Add that loss to the risk already open across related positions.
- Confirm that the holding period and event exposure fit your plan.
- Write one sentence explaining why this position belongs in the account now.
- Reject the order when you cannot make that case without changing your rules.
The final step matters. Rejection does not mean the model made an obvious mistake. It can mean the setup is valid and the account-level consequence is wrong.
That is the logic behind rejecting a valid but oversized order. The signal may remain technically defensible while the size makes the trade unsuitable.
Review the meaning before approving the number
NASA’s investigation did not conclude that calculation itself was useless. It exposed the cost of allowing a number to pass through a system without confirming that every component interpreted it the same way.
Before approving a queued trade, perform the trading equivalent of a unit check. Ask what the proposed position means for the whole account after execution, not merely whether the signal logic completed successfully.
Record the signal, proposed size, stop loss, account exposure, approval decision, and reason. Over time, that journal separates model quality from decision quality. It also makes rejection measurable. You can later review whether declined trades would have helped or harmed the account without pretending you knew the outcome in advance.
The next time every field is green but the resulting position feels wrong, stop before the order reaches the market. Find the mismatched assumption. A clean execution log cannot make an unwanted position acceptable.
Educational content, not financial advice.
Comments
No comments yet.