emperor-mew/voidly-causal-forest-hte-v1

Voidly Atlas Causal Forest HTE v1

Version: v1 | License: CC BY 4.0

Heterogeneous treatment effects of upcoming elections on censorship risk.

Headline finding

ATE: +9.6 percentage points lift in 7-day risk in the 30-day window before an election.

Honest caveats

• Heterogeneity is wide — some regimes show +30pp (Venezuela, Belarus) while stable democracies show ~0pp.
• Election event metadata is hand-curated from Wikipedia + GDELT — coverage gaps in 2026 Q1 will shift the ATE.
• Confounders not exhaustively addressed (no instrument); use as descriptive HTE, not causal effect estimate for production decisions.

Citation

@misc{voidly_voidly_causal_forest_hte_v1,
  title  = {Voidly Atlas: voidly-causal-forest-hte-v1 (v1)},
  author = {Voidly},
  year   = {2026},
  url    = {https://huggingface.co/emperor-mew/voidly-causal-forest-hte-v1},
  note   = {Open censorship-research ML stack. CC BY 4.0.}
}

Method foundation: Athey & Wager 2019 — Generalized Random Forests