Persistence and volatility: measuring whether states hold or churn.

The blind spot they cover

An average can describe a month that never actually happened.

Average gate rank is a mean, and means hide shape. A month spent entirely in Recovery and a month alternating violently between Growth and Floor can produce identical averages while describing completely different experiences — the first steady and predictable, the second a system unable to hold any state long enough for its deployment posture to mean anything. Persistence and volatility exist to make that difference visible. They're shape metrics rather than position metrics, and the build sessions classify them accordingly: pressure, health, and compression define what state you're in; volatility, transitions, and persistence describe what kind of month it was.

Persistence

Durability — how long gate states actually held.

Gate persistence measures holding power: the durability of the structural state across the month, with higher scoring stronger. High persistence means the system settled — whatever gate it occupied, it occupied it long enough for the deployment posture to be coherent and for the operator to trade a stable set of authorizations. Low persistence means states kept dissolving before they established anything. The build notes called this one of the most powerful structural metrics in the architecture, and the reason is that persistence is a precondition for almost everything else working: tier authorizations, pool budgets, and recovery arithmetic all assume a state that lasts longer than the cycle deploying inside it.

Volatility and transitions

Oscillation, sized against the month itself.

The gate volatility index measures how unstable the structural regime was relative to the size of the month, with lower scoring stronger — the normalization matters, because raw switching counts penalize longer months unfairly. Alongside it, gate transition count is the blunter companion: simply how often the system moved between gate states, again lower-is-stronger. The two overlap deliberately and diverge usefully. A month with many small transitions around a boundary scores differently from one with fewer but violent swings across multiple bands, and the pair distinguishes ordinary boundary noise from genuine structural instability.

• High transitions with high persistence is a boundary-hugging month — noisy but not unstable.
• Low persistence with high volatility is the genuine instability signature, and it routes to the Instability regime by design.
• Transitions are normalized against month size in the volatility index precisely so short months aren't flattered.

Why churn is expensive

Instability taxes the system even when the average looks fine.

A whipsawing month costs more than its average rank suggests, for reasons that are structural rather than psychological. Deployment authorizations change with gate state, so a system crossing boundaries repeatedly is re-pricing its pool rows constantly — and because gate transitions bind on the next cycle rather than mid-cycle, the account spends much of the month deploying against a state it has already left. Recovery arithmetic suffers similarly: climbing out of a defensive gate requires sustained deployment at reduced tiers, and a system that keeps falling back never accumulates the consecutive progress recovery depends on. This is why the regime engine treats low persistence and high volatility as genuine negatives rather than curiosities: churn is a real cost, paid in compounding that never happens.

The key idea

Shape is a dimension, not a detail.

Three of the six regime inputs measure position and three measure shape, and the split is the framework's most underrated insight. Position tells you what the month was; shape tells you whether it was a month at all or twenty separate ones. An operator reading only averages will consistently misjudge unstable periods as ordinary — and the instability is precisely what makes the next month harder. Persistence and volatility are how the diagnostic layer refuses to let that hide.