Simulation controls

Simulation speed: 1.0×

Feature-effect strength

Engagement effect: 1.00× Compensation effect: 1.00× Manager-quality effect: 1.00×

Each slider scales the true generative Cox coefficient (β) used to draw departures: hazard = h₀·exp(β₁·strength₁·z_engagement + β₂·strength₂·z_comp + β₃·strength₃·z_manager). Raising a slider makes that feature drive risk more strongly for every future instant.

Kaplan-Meier subgroup

Cohort stats

Sim week0

Active employees—

Events (departed)—

Censored—

Median survival (KM)—

Hazard ratios (per +1 SD)

Engagement — true—

Engagement — estimated—

Compensation — true—

Compensation — estimated—

Manager qual. — true—

Manager qual. — estimated—

"Estimated" values come from a real partial-likelihood fit (gradient ascent on the Cox log partial likelihood, Breslow tie handling) run on the cohort's actual simulated event/censoring times — not the generative truth. Compare them to see estimation converge as more events accumulate.

Employee feature space — engagement × compensation × tenure (drag to rotate, scroll to zoom). Colour/size = current hazard; dark red = departed; blue = censored at horizon.

Kaplan-Meier retention curve — selected subgroup, computed from real event/censoring times