A mechanical trading system is defined by its completeness: every entry, every exit, and every pause is specified before the session opens. The Operator desk examines what happens in the gap between that specification and the behaviour of the person running it — a gap that widens most reliably under losing-streak conditions, which are precisely the conditions the system was built to survive.

This piece covers the mechanics of mid-run intervention: what changes when a rule is overridden, how that change propagates through the system's expectancy, and why the cost is structural rather than incidental. It does not address any specific market or instrument. The mechanism is the same whether the system trades equities, futures, or spot currency.

How a Rule Gets Overridden, Step by Step

A mechanical system generates signals from a fixed set of conditions. When those conditions are met, the system places an order at a defined size and holds the position until a predefined exit condition is triggered. The operator's role during a live run is execution only — confirming fills, monitoring for data errors, and logging results. No discretionary input is specified at any stage.

Intervention begins when the operator substitutes a real-time judgment for one of those pre-specified actions. The most common forms are: closing a position before the exit condition is met ("early exit"), declining to take a signal that the system has generated ("skip"), and adding size to an open position outside the system's position-sizing rule ("override size"). Each of these is a different operation, but all share the same structural property — they replace a rule-derived action with a judgment-derived one.

When an early exit is taken, the system's recorded outcome for that trade is the operator's exit price, not the price the rule would have produced. The next time the system's historical expectancy is reviewed, that trade appears in the log as a system trade. It is not. The log is now a mixture of rule-generated outcomes and judgment-generated outcomes, and the two are indistinguishable after the fact unless the operator keeps a parallel annotation record.

When a signal is skipped, the system's win-rate denominator is unchanged in the operator's mind but the numerator is selectively filtered. Operators disproportionately skip signals during drawdown periods, which are statistically the periods most likely to precede a recovery in a system with positive expectancy. The skipped trades are not random omissions; they are correlated with the condition the system was designed to trade through.

Override sizing — adding to a position beyond the system's rule — changes the risk profile of a single trade without changing the system's stated parameters. If the trade resolves against the position, the loss is larger than any historical simulation would have recorded for that signal. The system's drawdown statistics, which were computed on fixed sizing, no longer describe the actual drawdown the operator experienced.

What the System Needs to Be True — and Cannot Make True

A mechanical system's expectancy is computed over a sample of trades generated under consistent conditions. For that expectancy to remain a valid description of the system's forward behaviour, three conditions must hold: the market regime that produced the historical sample must persist, execution must be consistent with the simulation's assumptions, and the operator must follow the rules on every signal — not on a selected subset.

The third condition is the one the system cannot enforce. A backtested rule set has no mechanism for requiring adherence. It produces a number — expected value per unit of risk per trade — that is valid only if the full trade distribution is realised. Partial adherence produces a different distribution. The system cannot detect the difference from inside its own logic.

The first condition — regime persistence — is also outside the system's control, but it is at least observable. Volatility measures, correlation structure, and liquidity depth can be monitored and compared against the parameters of the historical sample. Operator adherence cannot be monitored by the system at all; it requires the operator to maintain an honest annotation log, which is itself a discipline problem.

There is an asymmetry worth noting: operators tend to follow rules precisely during winning streaks and deviate during losing streaks. This means intervention is not randomly distributed across the system's trade history. It is concentrated in the draws, which are the trades that most determine whether the system's risk management holds. A system that is followed only when it is winning is not being tested; it is being harvested selectively, and the harvest is finite.

The Round-Trip Floor and the Hidden Cost of Deviation

Every trade a mechanical system places carries a round-trip cost composed of the exchange commission, the bid-ask spread at the time of execution, and slippage — the difference between the signal price and the actual fill price. On a centralised exchange with a maker-taker fee structure, a passive (maker) limit order typically incurs a lower commission than an aggressive (taker) market order. As a reference point, Coinbase Advanced Trade publishes a maker fee schedule that begins at 0.40% per side for low-volume accounts and falls toward 0.00% at the highest volume tiers; taker fees follow a similar but higher schedule. These figures are per-trade and per-side, so the round-trip cost at the entry tier is at minimum 0.80% of notional before spread or slippage is counted.

A system clears this floor through one of three mechanisms: frequency (many small edges compound above the cost), size (a single large position where the edge exceeds the cost as a percentage), or holding period (a longer hold where price movement dwarfs the entry and exit cost). The system's design specifies which of these it relies on. Intervention disrupts the chosen mechanism directly.

An early exit on a holding-period system cuts the trade before the price movement that was intended to dwarf the round-trip cost has had time to develop. The commission and spread are still paid in full. The edge that was meant to cover them is not realised. The trade records a cost without the compensating return the system's logic required. Across a run of early exits, this produces a persistent drag that does not appear in the system's backtested results because those results assumed the exits occurred at the rule-specified price.

Skip decisions carry a different cost that is harder to quantify: the cost of the trades not taken. If the system has positive expectancy and a signal is skipped, the foregone expected value is a real cost even though no commission is paid. This cost does not appear in any fee schedule. It appears only in the divergence between simulated and realised performance over the same period.

The Structural Breakdown: When Discretion Consumes the Edge

The terminal failure mode of a mechanical system is not a bad signal or an adverse market move. Both of those are within the system's design envelope — they are the losses the system's expectancy calculation already accounts for. The structural failure is the gradual replacement of the rule set with the operator's judgment, a process that typically occurs incrementally and is invisible in the trade log unless annotated separately.

The failure is structural because it is self-reinforcing. Each intervention that happens to produce a better outcome than the rule would have — and some will, by chance — provides evidence, in the operator's experience, that judgment improves on the system. Each intervention that produces a worse outcome is more easily attributed to bad luck than to the act of intervening. The asymmetry in how outcomes are attributed means the rate of intervention tends to increase over time regardless of whether the interventions are net beneficial.

At the point where the majority of exits are discretionary, the system has ceased to exist as a mechanical system. What remains is a discretionary process that uses a mechanical signal for entries only. That is a different strategy with different risk properties, different expectancy, and no historical simulation to validate it. The operator is running a system they have never tested against the rules they are actually following.

There is no threshold intervention rate below which the system is unaffected. Even a single skipped signal changes the realised distribution. The practical consequence is that a system's performance record, once contaminated by unlogged intervention, cannot be used to evaluate whether the underlying rules have positive expectancy. The record describes a mixture of two strategies, and the mixture's properties cannot be recovered without complete annotation of which trades were rule-generated and which were judgment-generated.

This is the cost the title names: not a fee, not a spread, but the destruction of the evidentiary basis on which the system was justified in the first place. A system that has been intervened upon without annotation is a system whose expectancy is unknown. Running it further accumulates cost — commissions, spreads, slippage — against an edge that can no longer be verified to exist.

The gap between a rule and following it is not a character flaw in the operator; it is a predictable property of any process that requires consistent action under conditions designed to produce inconsistent emotion. Mechanical systems do not close that gap — they make it visible, trade by trade, in the annotation log.

Note: This explains how a process works. It is not legal advice, it is not specific to any debt, and it is not a substitute for a licensed attorney in your state. Rules and time limits vary by state and change over time — check the cited sources.