Six metrics, one composite: how a month becomes a regime.

The six inputs

Three that define the state, three that shape it.

The engine reads six monthly metrics from the gate/brake telemetry, and the build sessions split them conceptually. The core determinants — Gate Pressure (how heavy the month was from drawdown, spike clustering, and compression), Structural Health (how well the system handled the month, not merely whether it was easy), and Gate Compression (constraint versus capital freedom) — define what state you're in. The shape modifiers — Gate Volatility Index (instability relative to month size), Gate Transition Count (how often gate states switched), and Gate Persistence (how durably states held) — describe what kind of month it was. Pressure plus health plus compression tells you the state; volatility plus transitions plus persistence tells you its character.

The scoring ladder

Every metric contributes through the same −2 to +2 scale.

Each metric is scored against fixed thresholds onto a five-level ladder: +2 strongly positive, +1 mildly positive, 0 neutral, −1 mildly negative, −2 strongly negative — with direction respecting each metric's own polarity, since lower is stronger for pressure, volatility, and transitions while higher is stronger for health and persistence. The six scores then sum to a composite ranging from −12 to +12, and that single number drives the main regime assignment. The non-negotiables around this are explicit: no weights in the first version, fixed thresholds rather than z-scores or rolling percentiles, and six metrics only — do not turn the classifier into a twenty-metric score monster.

The override conditions

Two readings that outrank the composite arithmetic.

The classification logic isn't purely additive — two conditions short-circuit it. A very high compression reading routes the month to Drawdown Control outright, and a high-but-lower compression reading forces a choice between Compression and Instability depending on the composite. The reasoning is priority hierarchy: severe capital constraint is a state you're in regardless of how the other five metrics happen to average out, and letting a good persistence score paper over active containment would produce exactly the false-positive classifications the engine exists to prevent. It's the same doctrine as the authority stack — capital-state conditions outrank aggregate assessments — implemented at the classifier level.

Why fixed and unweighted

A first production version earns its complexity later.

The design rules read as deliberate under-engineering, and each has a reason. Fixed thresholds rather than adaptive statistics: the engine must mean the same thing in month three as month thirty, and adaptive thresholds quietly redefine what a regime is as history accumulates. Equal weights: weighting requires evidence about which metric predicts what, and that evidence doesn't exist until months of classifications can be compared against outcomes. Six metrics: every added metric multiplies threshold-calibration surface and dilutes the composite's interpretability. The documented future path is exactly the MARS pattern — empirical threshold calibration using six to twelve months of history, a confidence score showing whether the six metrics agree, a contradiction check against SDE and CP3 states — all of it earned rather than assumed.

• One main regime per month, one sub-label. No stacking, no hedged classifications.
• The compression-score direction carries a standing QA flag — its polarity must be audited before live reliance.
• Fixed thresholds mean regime history stays comparable across the entire record. That comparability is the point.

The key idea

The engine's job is cognitive compression, not more numbers.

Six telemetry metrics, individually, are what an operator already has and can't act on — 'pressure is 62, stability is 71, three transitions.' The composite collapses that into one sentence: the system is in Compression. That's the whole contribution, and the design restraint is what protects it. Every additional metric, weight, or adaptive threshold would buy marginal accuracy at the cost of the one property that makes the output useful — that it's a single, stable, comparable state you can actually operate against.