Month three. Year one. Year three. And the operator for whom nothing changes.

Layer 01 — Why a trajectory and not a promise

The variable the system controls is decision quality, not market outcome.

MARS acts on what the operator records, how risk is deployed, which findings are produced, and which changes are allowed into production. It does not act on whether markets cooperate. That boundary is what makes a trajectory describable at all: the improvements are in process quality, they accrue on a schedule set by evidence accumulation rather than by effort, and they are individually verifiable. What cannot be described is the return, because the return depends on a variable no framework governs.

Layer 02 — Month three

Behaviour changes first. The analytics are still nearly useless.

The earliest returns come from the parts that require no history: deployment is authorised rather than felt, drawdown routes tier ceilings automatically, and open exposure is a figure rather than an impression. Oversizing after a good week simply stops being available. Meanwhile the analytical layer has almost nothing to say — sample sizes are too small for expectancy to be distinguishable from noise, and the benchmark has not been fitted to a real branch mix. This asymmetry is the least advertised fact about early adoption and the most useful one to expect: the boring layer pays immediately, the interesting layer does not.

Layer 03 — Year one

Attribution becomes possible, and the first uncomfortable findings arrive.

With four quarters of consistently classified evidence, per-branch expectancy separates from account totals — which branch carries the account, which is subsidised, whether an accelerator is earning its variance. Adherence has enough observations to show a pattern rather than an incident, and the benchmark has been calibrated to the actual mix, so a stretch can finally be classified as inside or outside the envelope. Most operators meet their first genuinely unwelcome finding somewhere in this window, and it is usually about behaviour rather than about method.

Layer 04 — Year three

Drift becomes visible, and the record starts answering questions faster than they can be asked.

Slow deterioration is invisible over any short window by definition — it is the failure mode that looks like variance until it is expensive. Three years of stable branch definitions makes it a measurable trend line, and separates it from the ordinary bad quarter. The promotion history matters at this stage too: the current rules stop being a configuration and become a set of decisions with recorded reasons and recorded results. This is where the compounding actually is, and it cannot be purchased, accelerated, or supplied by a higher tier.

Layer 05 — The flat case

For one operator the curve never leaves the floor, and it is predictable in advance.

The trajectory assumes faithful capture. An operator who logs sporadically, reconstructs entries at weekends, or stops updating during drawdowns receives the month-three benefits indefinitely and never the others — because every downstream layer inherits the record, and a record with its worst stretches missing overstates expectancy, understates drawdown, and misdates every structural finding. This is not a warning about discipline in the abstract. It is a description of the one input that determines whether the rest of the curve exists.

• Sporadic capture caps the value at the month-three layer, permanently
• Missing stretches are directional error, not random noise
• The failure is silent — every reading still renders, and still looks correct