A rule is a promise. A rail is a structure.

The distinction

One asks for compliance. The other removes the steering wheel.

A rule is a commitment an operator makes to themselves and re-honours on every occasion it applies — which means it is tested precisely as often as it matters, and by a person whose state varies. A rail is a structure the workflow physically runs on: the sizing figure arrives already computed, the gate state is already set, the directive already says what may be deployed. Departing from a rail is not a matter of failing to honour it. It requires deliberately doing something else, and that difference in the shape of the failure is the entire argument. The difference shows up most clearly in how each fails: a broken rule leaves no trace and is often not remembered by the following week, while a departure from a rail leaves an artifact somebody can read afterwards.

Path of least resistance

The design goal is that following the system is the easy option.

The mechanism is not enforcement, since a spreadsheet cannot prevent anything. It is friction asymmetry. Following the rail requires reading a number that is already there; departing from it requires overriding a mode, entering a figure the system did not produce, and logging a reason. Neither is impossible. One is effortless and one is not, and behaviour under pressure follows effort gradients far more reliably than it follows intentions. The system does not stop the operator — it makes the compliant path the lazy path, which turns out to be enough most of the time.

Why rules fail predictably

They are tested most heavily in the state that breaks them.

The failure mode of a rule is not random. Rules are challenged specifically during drawdown, during winning streaks, and during fatigue — the three states in which the operator's judgement is least reliable, and where a rule's only defence is that same judgement. A rule is therefore weakest exactly where it is needed most, and no amount of conviction at the moment of writing changes that. This is not a claim about weak-willed traders; it is a structural property of any control whose enforcement mechanism is the thing being controlled.

What makes something a rail

Three properties, and all three are required.

The term is not a synonym for a strict rule. A rail must satisfy all of the following, and something meeting only two is a rule wearing better language:

• Computed, not recalled — the value arrives from the system, not from memory.
• Present at the decision — it is in front of the operator when the choice is made, not in a document.
• Departure leaves a record — an override is visible afterwards without anyone having volunteered it.

The honest gap

A workbook cannot physically stop anything.

MARS V1 is an Excel system, not a broker integration, and a determined operator can type any number into any platform regardless of what the throttle says. The rails are evidentiary rather than physical, and the design accepts this openly. The bet is that most discipline failures are not defiant — they are quiet, ad hoc, and unrecorded, and they do not survive being written down first. A trader who logs an override with reasoning has already done most of the thinking the rail was there to force, which is a large fraction of the benefit even when the override proceeds. The evidentiary model also has an advantage the physical one lacks — it keeps the operator responsible for the decision, which means the record produced is a record of judgement rather than a record of what the software permitted.

Rails and the version ladder

Later editions can tighten the mechanism, not change the doctrine.

The distinction between evidentiary and physical enforcement is a property of the current implementation rather than of the doctrine. A cloud edition can hold state the operator does not control and refuse to display a size the gate does not authorise. That would strengthen the rails without altering what a rail is — the three properties above remain the definition, and a future version that enforced harder while abandoning the record would be a worse rail, not a better one.