109 tabs, twelve zones, one question.

Layer 01 — The two-file split

The workbook configures. The engine computes. Neither writes over the other.

The Lab is deliberately two cooperating pieces. The workbook is the configuration surface and the report: every assumption is set there, and every result is written back there as report-grade tables. The engine is a separate Python process that reads the configuration, runs the simulation in numpy, and writes results into a copy of the workbook rather than into the workbook itself. The original is never modified by a run, which is what makes a run repeatable rather than a one-way transformation of your own settings.

Layer 02 — The golden sequence

Recalculate, run, recalculate.

The engine reads cached formula values, not formulas. Saving a workbook can drop those caches, and an engine reading stale caches produces a flat account or a zero-trade run that looks like a modelling failure and is actually a housekeeping failure. So the operating habit is fixed: recalculate the input workbook before the run so configuration formulas are evaluated, then recalculate the output workbook after the run so dashboards pick up the values the engine wrote. Blank dashboards after a run are almost always this, and almost never anything else.

• Blank Zone 10 table with a Set Full note: the detail level was not Full. Not a defect.
• Zero trades or a flat account: the input workbook was saved without a recalculation.
• Blank comparison tabs: the output workbook was never recalculated once.

Layer 03 — Detail as a speed control

Summary, Standard, Full: three sizes of the same run.

Output Detail Level decides how much of the output suite the engine computes. Summary populates Tables A through F only. Standard adds G through S, the core benchmark suite. Full adds T through AF, the advanced attribution and stress tables including branch attribution, variant attribution, fat-tail dependency, fee stress, the risk spectrum, gate transitions, throttle value-add, override impact, and run confidence. The tables that answer why a result happened live almost entirely in Full, which is why a marginal result should always be re-run at Full before it is acted on.

Layer 04 — The truth anchor

The calibration bridge stops the exit mechanics from inventing an edge.

Zone 7 pulls measured expectancy from Compliance Panel 3, weekly branch hit rates from the Scorecard, execution efficiency from the MAE/MFE Lab, and week-by-week actuals from the live account. When CP3 calibration is active, the engine will not let the exit structures manufacture a blended expectancy higher than the measured edge. That single constraint is what separates a credibility tool from a wishful one, and it is why the Lab is willing to report that a risk doctrine the operator likes is unsustainable at the edge the operator actually has.

• Enter only the exit levels you genuinely trade in Exit_Calibration; the engine interpolates the rest of the ladder.
• Variant usage is entered as monthly trade counts, not abstract weights, and read as a mix ratio rather than a quota.
• Data_Quality_Check flags missing data, invalid labels, and insufficient evidence before a run is permitted.

Layer 05 — What it will not tell you

The Lab models the assumptions you give it.

It stress-tests a plan; it does not validate one. Feed it an optimistic edge and it will faithfully simulate an optimistic account, in convincing detail, across fifty thousand paths. Its value is not that it predicts. Its value is that it exposes fragility: where a thin edge, an aggressive risk band, or a fee-heavy cost model quietly converts a system the operator believes is good into a lockout machine.