Open methodology summary. Algorithms run entirely in your browser; this page describes the research intent, not a clinical claim.
Digital interactions are modeled as a graph where stress proxies (sprawl, switching, repeated returns) play a role analogous to a percolation parameter. Phase-like transitions are heuristics for research visualization, not physical law claims about cognition.
Density-based spatial clustering of applications with noise (DBSCAN) separates dense behavioral modes from sparse "noise" periods that may correspond to spawning unrelated sub-problems. Eps can be auto-tuned via silhouette score over a candidate grid.
Time bins are nodes. Edges combine temporal adjacency, shared cluster membership within a lag window, and file co-occurrence (Jaccard overlap). Weights incorporate a sunk-cost proxy from repeat depth within each bin.
Ground-truth crisis dates are user-annotated; predictive statistics require prospective studies and IRB oversight. N=1 designs (e.g., dense self-tracking) can complement population work when hypotheses are pre-registered.