Operator brief · 169

The research wall, implemented as an immutability guarantee.

The key idea

The split

One file holds intent; the other holds consequences.

The Lab is deliberately two cooperating pieces. The workbook is the configuration surface and the report format — every assumption is set there, every table is defined there. The engine is the simulation core: it reads the configuration, runs the paths, and writes results back as values into a copy. The copy is where the numbers land. The original is never touched. That sentence is doing far more governance work than its plainness suggests, because it means the artefact holding what the operator declared cannot be retroactively edited by the process that produced the results — not by accident, not by a crashed run, and not by an operator who would rather the inputs had been slightly different.

Why immutability matters here

A research record is only evidence if it cannot be quietly revised.

The characteristic failure of spreadsheet-based research is not fraud; it is drift. An assumption gets nudged during a review, the workbook is saved, and three weeks later nobody can reconstruct which configuration produced the chart in the memo. The result outlives the inputs and becomes free-floating — cited, trusted, unreproducible. The two-file split closes that path structurally. The configuration workbook remains the canonical statement of intent; each run generates a separate, timestamped consequence. If a later reviewer wants to know what was assumed, the answer exists as a file rather than as a recollection, and it exists in a form the run itself could not have altered.

FigureThe research loop — the source workbook is read from, never written to

Consequences accumulate as separate artefacts while intent stays in one canonical place. The loop closes when a reading motivates a configuration change, which is itself a declared act rather than a side effect of running.

The disposability that follows

Run artefacts are cheap on purpose, and that is what makes exploration free.

Because results land in a copy, a run costs nothing but time. There is no reason to hesitate before testing a hostile assumption, no reason to preserve a disappointing output, and no reason to feel that a bad run has damaged anything. This is the practical half of the boundary and it is easy to underrate. A sandbox where experiments feel expensive stops being used, and an unused sandbox pushes exploration into the only remaining venue, which is the live account. Cheap, discardable, reproducible artefacts are not a convenience feature. They are the mechanism that keeps curiosity from migrating somewhere it can cost money.

What still requires a person

The file rule prevents contamination; it does not confer permission.

It is worth being precise about the guarantee's limits, because architecture can create false confidence as easily as it creates safety. Immutability of the source workbook means a run cannot silently change what was assumed. It does not mean a compelling run cannot change what the operator does next — that path remains wide open, and it runs through the promotion workflow rather than through the file system. Nothing about the two-file split prevents an operator from reading a flattering distribution and deciding to size up tomorrow. The wall that stops that is doctrinal and lives elsewhere. What the split guarantees is narrower and still valuable: whatever happens next, the record of what was tested survives intact.

The overrides that get logged

Command-line arguments do not escape the record — they enter it.

Run parameters can be overridden at execution time without editing the workbook: path count, seed, and the output detail level can all be supplied directly to the engine. This is a genuine convenience and it opens a genuine hole, since a run could otherwise differ from its stated configuration with no trace. The Lab closes it by logging the overrides in the run note, so the artefact records what was actually executed rather than what the workbook merely said. The principle generalises across the whole system: a flexible interface is safe exactly to the degree that its flexibility is recorded, and unsafe the moment convenience produces an unlogged difference between the plan and the run.

The key idea

Prefer boundaries that hold when nobody is watching.

Governance written as a rule requires a person to remember it under exactly the conditions that make remembering hardest — late, invested, and looking at a number they want. Governance written as architecture does not. The two-file split is a small instance of a principle the whole rail is built on: wherever a boundary can be made structural, make it structural, and reserve human discipline for the decisions that genuinely require judgement. The research layer earns its freedom to explore anything precisely because the thing it explores from cannot be damaged by the exploring.

Connected inside MARS

Monte Carlo Lab →

Scenario Manager →

Prime Directive →

Every brief documents the same shipped system.

The complete MARS package — eleven workbooks, three TradingView indicators, the full manual library — $497.

Get MARS Pro — $497 Explore the Platform