for each time t:
  p_em(t) = (# of simulations with _k <= t) / N
# MFPT and hazard
MFPT = mean({_k finite})
compute hazard h(t) from empirical F_T
# Actor probabilities at horizon t:
for each actor i:
  p_i(t) = (# of sims with _k <= t and actor i selected at _k) / (# sims with _k <= t)
Outputs: pem(t)p_{\text{em}}(t), MFPT, hazard curve, actor shares pi(t)p_i(t). Also return ensemble percentiles for H(t),P(t),T(t)H(t),P(t),T(t).
4) Calibration of HcH_c and mapping to actors
Calibrate HcH_c using historical "emergence" episodes (e.g., prior leader surges) by mapping back to the composite index at known emergence dates. Choose HcH_c as the 75--90th percentile of historical pre-emergence values to avoid excessive false positives.
Candidate actor roster A\mathcal{A}: populate from political actors with plausible reach (e.g., national politicians, regional leaders, celebrity politicians, movement organizers). For each actor, define si(t)s_i(t) from data streams (social, organizational, financial). Update si(t)s_i(t) in near-real time.
Soft attribution versus hard label: The model returns probabilities for multiple actors. Do not pick a single "winner" until pip_i exceeds a high threshold (e.g., pi>0.6p_i>0.6) and H(t)H(t) crosses HcH_c in multiple ensemble members.
