Monthly, and non-negotiably so: the unit the regime engine runs on.

The rule

Keep the engine monthly — first of the non-negotiable design rules.

The manual's design constraints open with it: the engine is a month-by-month classification of internal system performance, and the cadence is not a parameter. That is stronger language than the other rules receive, and the reason is that three of the six input metrics are only meaningful over a span. Gate Transition Count measures how often the system moved between capital states; Gate Persistence Score measures how long those states held; Gate Volatility Index measures how unstable the structural regime was relative to the size of the period. Each of these is a statement about behaviour over time rather than a level at a point. Run them over a week and two of the three frequently have no variation to report — a week in which the gate never moved yields a transition count of zero and a persistence score that says only that nothing happened.

Against the week

The scorecard's unit is right for the scorecard and wrong here.

The weekly unit earns its place in the expectancy layer because trades accumulate fast enough for a week to carry a usable sample and slow enough that a week is not noise. The regime engine's raw material accumulates far more slowly. Gate states are supposed to persist — that is what a capital-state ladder is for — so a system behaving correctly will produce very few transitions in any given week. Measuring persistence over a period shorter than the thing being measured typically lasts produces a reading dominated by whether a boundary happened to fall mid-state. The month is not a compromise between responsiveness and sample size. It is roughly the shortest window in which gate behaviour has enough events to have a shape.

• Transition count, persistence, and volatility are all span metrics — they need multiple state changes to say anything.
• A weekly regime reading would classify boundary placement as often as it classified system condition.
• Pressure and structural health would survive a weekly cadence; the other four would not.

The cost, accepted

A month is slow, and the engine is not the fast layer.

The cadence has a real drawback: a system that begins deteriorating in the first week of a month will not receive a regime label reflecting it until the month closes, and by then the deterioration has had four weeks to run. This is a genuine lag and it is deliberately tolerated, because the regime engine is not the component responsible for catching fast problems. The gate reacts to drawdown continuously and holds hard authority precisely so that nothing has to wait for a monthly reading. The weekly scorecard catches expectancy problems at weekly resolution. The SDE's rolling windows catch structural drift between the two. The regime engine sits above all of them answering a slower question — what kind of month was this, in governance terms — and a slow question answered slowly is not a lag. It would only be a lag if something urgent depended on it, and by design nothing does.

What the month makes possible

Twelve labels a year is a readable history.

There is an underrated benefit to the slower unit. Twelve regime labels across a year is something an operator can hold in their head and read as a narrative — three months of Balanced, a Transitional month, two of Compression, then Fragile Recovery — and that narrative is genuinely diagnostic in a way fifty-two labels would not be. A weekly cadence would produce a series long enough to require its own summarisation layer, which would reintroduce exactly the compression problem the engine was built to solve. The month is the largest unit that still gives enough resolution to see a change, and the smallest that produces a history short enough to be read directly.

The key idea

The measurement window has to be longer than the thing being measured.

Gate states are meant to hold. Persistence, churn, and volatility of a slow-moving layer cannot be observed on a fast cadence, and forcing them onto one produces numbers that describe the calendar rather than the system. The month is not a preference about review rhythm. It is the shortest window in which the engine's own inputs mean anything.