Voidly Atlas Bayesian Corroboration v1

Version: v1 | License: CC BY 4.0

Multi-source Bayesian evidence fusion. Given conflicting signals from OONI, IODA, CensoredPlanet, and Voidly probes for a (country, domain, day), returns a calibrated posterior probability that the observation is real censorship.

Eval

Metric Value
roc_auc 0.9200

Honest caveats

  • AUC 0.92 reflects evaluation on retained labels — model is a fusion layer over evidence, NOT a primary classifier.
  • Source-conflict weights are hand-prior'd; full hierarchical Bayes is future work.

Citation

@misc{voidly_voidly_corroboration_v1,
  title  = {Voidly Atlas: voidly-corroboration-v1 (v1)},
  author = {Voidly},
  year   = {2026},
  url    = {https://huggingface.co/emperor-mew/voidly-corroboration-v1},
  note   = {Open censorship-research ML stack. CC BY 4.0.}
}
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support