The composite score
Each signal on the panel sits in one of four states: ok, watch, warning, or critical. The composite score is a weighted sum of the active states — watch contributes half a point, warning one point, critical a point and a half, each multiplied by that signal's own weight in the panel.
From score to risk band
The score maps onto four bands calibrated against 1967–2026 history: low (under 5, where measured recession frequency sits below the ~16% unconditional base rate), elevated (5–9), high (9–13), and severe (13+). The band boundaries were calibrated against the historical record rather than fitted — the band, not the raw number, is the estimate to read.
Why a sum, not a model
A fitted model (logit/probit) can squeeze more precision out of the same panel — but its coefficients are opaque and unstable out of sample. The weighted sum is fully auditable: every point of the score can be attributed to a specific signal, which matters more for a public monitor than fractional accuracy gains.