7) Decision rules (examples)
Watchlist: If pem(7d)>0.15p_{\text{em}}(7d) > 0.15 AND top actor pi(7d)>0.25p_{i^\ast}(7d) > 0.25, activate targeted engagement (dialogue channels) with actor's known networks and tighten transparency around grievances they cite.
Escalation alert: If pem(30d)>0.4p_{\text{em}}(30d) > 0.4 OR MFPT < 14 days, declare high-risk and activate contingency plans (accelerated U, mediated talks, surgical security posture).
Attribution caution: Only label an actor a "likely black horse" when pi>0.5p_i > 0.5 and H actually crosses HcH_c in multiple independent data windows (to avoid false-positive stigmatization).
8) Mapping method to concrete actors (practical note)
Use actor-level features (social reach, network centrality, rhetorical alignment) to score si(t)s_i(t).
For a given actor aia_i, compute a mobilization potential:
si(t)=1norm(follower_growthi)+2norm(mention_connectivityi)+3norm(organ_linksi)s_i(t) = \sigma_1\cdot\text{norm}(\text{follower\_growth}_i) + \sigma_2\cdot\text{norm}(\text{mention\_connectivity}_i) + \sigma_3\cdot\text{norm}(\text{organ\_links}_i)
Normalize and feed into the attribution step. The model's output is probabilistic---it should be used to guide monitoring and engagement, not to accuse.
9) Example (toy numbers, illustrative)
Ensemble N=2000; horizon 90 days; calibrated Hc=0.65H_c=0.65.
Output: pem(30d)=0.28, pem(90d)=0.47;p_{\text{em}}(30d)=0.28,\ p_{\text{em}}(90d)=0.47; MFPT median = 38 days.
Actor shares at 30d: Anies 0.18, AHY 0.12, Gibran 0.08, Puan 0.06, Dedi 0.04, others 0.52 (fragmented).
 Interpretation: moderate chance of emergence within 1--3 months; high fragmentation makes single-actor dominance unlikely in the short run.
10) Limitations and safeguards
Data bias: social media over-represents certain demographics; actor scores must be weighted by offline organizational signals.
Self-fulfilling risk: publicizing high pemp_{\text{em}} can itself catalyze mobilization; control access to sensitive outputs.
Ethics: map probabilities to engagement (dialogue, relief), not repression; avoid using forecasts to pre-emptively criminalize actors.
11) Implementation roadmap (practical steps)
a. Build data pipelines (social API, event database, polling/trust index, transaction monitoring).
