Compounding geometry: how expectancy becomes alpha.

The asymmetry

Losses and gains are not symmetric. Ever.

Lose 10%, and 11.1% recovers it. Lose 30%, and the requirement is 42.9%. Lose 50%, and you owe 100%. This convexity is why the entire MARS governance stack — gates, throttle, tiers — is oriented toward keeping drawdowns shallow: the same expectancy compounds dramatically faster when it never has to dig out of a deep hole. Defense is not the opposite of growth; in geometric terms, it is the mechanism of growth. The convexity is not a risk-management opinion; it is arithmetic, and it applies whether or not the operator agrees with it.

Acceleration

Slope says growing. Acceleration says how the growth is aging.

Equity slope confirms direction, but the second derivative carries the early warning: positive slope with fading acceleration means the machine is still climbing while losing thrust — often the first visible sign of edge decay or friction creep, weeks before the curve flattens. MARS tracks acceleration explicitly because operators who watch only the slope celebrate right up until the plateau. By the time a flattening curve is obvious, the acceleration reading has usually been negative for a considerable while.

Why the median matters more than the mean

Average outcomes are earned by paths most accounts never take.

A distribution of compounding outcomes is skewed by construction: a handful of paths run away upward and pull the average with them, while the typical path sits well below it. Planning against the mean therefore means planning against an outcome the median account does not experience. MARS reads the median and the adverse percentiles precisely because they describe the likely life of the account rather than its best available biography — and because a plan that only works on the flattering path is not a plan, it is a hope with arithmetic attached.

Frequency, edge, and the exponent

Three levers move the curve, and they are not equally safe to pull.

Growth responds to per-trade edge, to how often that edge is deployed, and to how much is risked each time — and the third lever is the one that also governs ruin. Increasing risk raises the exponent and the variance together, so a doubling of size does not double the outcome; it widens the distribution in both directions and moves the adverse tail toward depths the recovery arithmetic cannot repay. The safest lever is frequency at unchanged quality, the second is edge itself, and size is last precisely because it is the only one that can end the sequence rather than slow it.

• Frequency at constant quality: compounds without widening the tail.
• Edge improvement: the slowest lever and the most durable.
• Size: raises the exponent and the ruin probability together.

Time is the term the exponent multiplies

The only irreplaceable input is the one nobody can buy back.

Every other variable in the compounding expression can be improved. Edge can be sharpened through better selection and management; frequency can be raised by trading more markets or shorter horizons; even the base can be increased by adding capital. Time cannot be recovered once spent, and a sequence that ends early forfeits every remaining period regardless of how good the edge was. This is what places survival above optimisation in the MARS ordering, and it is why the governance stack accepts a lower theoretical growth rate in exchange for a much higher probability of still being running in five years. An account that compounds modestly and continuously beats one that compounds aggressively and stops, and the gap between them widens with every year that the first one keeps going.

The long game

Survival is the multiplier on everything else.

Compounding rewards the trader who is still standing in year five more than the one who was brilliant in month three. Every MARS discipline — sizing to the P10 path, compressing under drawdown, deploying on the evidence gradient — is a compounding decision wearing a risk-management costume. The formula A = P(1 + r)ᵗ only pays its exponent to accounts that protect the P.