The State-Change Signal: Graded, Uncertain, Never an Alert

The procedure

Detrend with a Gaussian kernel, bandwidth 10% of the series length.
In a rolling window of 50% of the series, compute lag-1 autocorrelation (AR1) and standard deviation.
Compute Kendall's τ for the trend of each indicator.
Generate 1,000 surrogate series from the best-fitting ARMA model, or AR(1). The one-sided p is the share of surrogate τ at least as large as the observed τ.
Repeat over a sensitivity grid of windows from 25% to 75% and a range of bandwidths. Robustness is the share of grid cells with p below 0.05.

Procedure from Dakos et al., 2012.

Rules

State-Change Signal rules
Rule	Default
Data floor	At least 100 equally spaced observations (more than three months of nightly data)
Label	"Consistent with slowing" only if p < 0.05 and robustness ≥ 70%; otherwise "not consistent" or "insufficient data"
Output form	A graded statement, for example "τ(AR1) = 0.42, p = 0.03, consistent in 70% of the sensitivity grid"
Irregular data	Momentary data must not be naively interpolated, because interpolation creates false autocorrelation

The mandatory disclosure

Every State-Change Signal output carries this disclosure: "At the level of one person, these signals are uncertain. In published studies they appeared before fewer than half of changes, and they appeared in about 25% of people whose state did not change."

Basis: Helmich et al., 2024, Nature Reviews Psychology. The tier is Model, with mixed evidence. The signal is a research estimate of change in self-reported wellbeing patterns, not a prediction of any health condition.

What the disclosure means in practice

Because true changes are uncommon, most positive signals in a group can come from people whose state did not change. The false-alarm calculator shows how many, for your own numbers.

Questions

Why is it never an alert?

Because at the level of one person the evidence is too uncertain, and an alert would imply a prediction ATS-1 does not make.

Sources

Keep reading

False-alarm calculatorFree calculator: see how many change signals come from people whose state did not change, given how common real changes are.Early-warning signals explainedWhat early-warning signals are, why recovery slows near a shift, what group studies found, and why one person's signals are uncertain.Resilience (recovery half-life)How ATS-1 measures resilience as recovery speed: an AR(1) model of nightly HRV gives a recovery rate λ = −ln φ and a half-life t½ = ln 2 / λ over 60 days.Evidence tiersThe ATS-1 evidence tiers (Established, Model, Hypothesis, Inspiration), their definitions, and the current register of claims.

Take the next step.

The ATS-1 State-Change Signal: rising autocorrelation and variance tested against surrogate data, reported as a graded statement.

False-alarm calculator Early-warning signals explained