Data-driven, outlier-robust dynamic-state thresholds

Derives a fixed_state_thresholds()-shaped list from an observed sniff_trajectory_dynamics() table, so the cut points adapt to the dataset instead of fixed constants. The growth cuts are placed k robust deviations (median absolute deviation) from the median, which resists outliers such as a single fast-growing young core; the bounded metrics use their median as the neutral split, and dormancy_share keeps the absolute majority rule (its distribution is usually too concentrated at zero to standardize).

Usage

data_state_thresholds(dyn, k = 1)

Arguments

dyn

A sniff_trajectory_dynamics() tibble; only its metric columns are read, so the classification it currently carries is irrelevant.

k

Number of robust deviations (scaled MAD) above/below the median for the growth cuts (default 1): emergence_growth = median + k*MAD and decline_growth = median - k*MAD over the living cores' growth_rate. Larger k widens the maturity band (fewer emergence / dormancy calls).

Value

A list shaped like fixed_state_thresholds(): emergence_growth, decline_growth (robust, from the centrals' growth), emergence_novelty (median novelty of the centrals), convergence_entropy (median destination entropy of the absorbed), and dormancy_share (the default majority rule). Any quantity that cannot be derived (too few points, zero spread) falls back to its fixed_state_thresholds() value.

Details

Use it as a drop-in for the thresholds argument; the metrics are threshold-independent, so one pass is enough to derive and re-classify: dyn <- sniff_trajectory_dynamics(flow), th <- data_state_thresholds(dyn), dyn <- sniff_trajectory_dynamics(flow, thresholds = th).

See also

fixed_state_thresholds(), sniff_trajectory_dynamics(), plot_trajectory_dynamics()