Noise, drift, or break: classifying the gap before touching anything.

Species one: noise

Variance — the gap that time reabsorbs.

Noise is deviation with no structure: a below-median month with normal drawdown, normal gate dwell, normal tier usage, and nothing in the execution record to explain it — because there is nothing to explain. The benchmark itself says a quarter of governed futures sit below P25 at any moment; someone has to be them, and sometimes it's you. Noise's signature is isolation and reversion: the placement wanders, the cross-checks stay clean, and adjacent periods don't rhyme. The correct response is the hardest one — nothing. No sizing change, no rule tweak, no catch-up aggression. A system adjusted every time variance shows its face stops being a system and becomes a mood with spreadsheets.

Species two: drift

The slow leak that never triggers an alarm.

Drift is directional deterioration too gradual for any single review to flag: fee drag creeping as trade frequency shifts, branch mix sliding away from the benchmark's assumed weights, adherence softening a percent at a time, capture efficiency eroding. Its signature is trend without event — each month individually defensible, the sequence unmistakable. Drift hides from monthly reads and shows up in rolling ones, which is why the comparison record's closed vocabulary matters: three consecutive 'underperforming' classifications are a trend even when no single month was alarming. Drift's response is diagnostic, not dramatic — trace which cross-check is trending, find the leak in execution or mix, and fix the process. The edge usually isn't going anywhere; something is taxing it.

Species three: break

The step-change that persistence confirms.

A break is a structural change in the relationship between the system and the market: results don't wander below the bands, they relocate there, and the cross-checks corroborate — drawdown behavior outside its distribution, defensive gate dwell far past benchmark norms, EV readings that fail their own statistical gates in the analytics stack. Breaks are the rarest species and the most over-diagnosed, because a break is emotionally available every time noise hurts. The protective rule: a break may only be declared through the evidence threshold covered in the next brief, never from placement alone. What a break is not: an excuse. Gate compression during the diagnosis is the ladder doing its job, and no suspected break authorizes trading outside it.

The classification discipline

Default to the least dramatic species the evidence permits.

The taxonomy comes with an ordering rule: noise is the default, drift requires trend evidence, break requires persistence plus corroboration. The burden of proof escalates with the drama of the diagnosis, because the costs are asymmetric in a specific direction — treating a break as noise costs you months of degraded results, but treating noise as a break costs you the system itself: rules abandoned mid-variance never get the sample size to prove they worked. Classification is also not a one-off: a gap filed as noise gets re-examined at the next review with one more period of evidence, and reclassification upward is always legal. What's illegal is skipping the ladder.

• One adverse month: noise until proven otherwise. Full stop.
• Three trending periods with a trending cross-check: drift — diagnose the leak.
• Persistent sub-band placement plus corroborating structural evidence: break protocol, through the formal threshold.

The key idea

The response inherits its correctness from the classification.

Every gap ultimately gets a response — hold steady, repair a leak, or re-examine the edge — and each response is only correct for its species. The entire value of the taxonomy is sequencing: classify first, with the benchmark's cross-checks as evidence, and the response mostly picks itself. React first, and the response is a coin flip wearing conviction. The comparison workflow locates the gap; this doctrine names it; and naming it correctly is the difference between governing a system and second-guessing one.