Probability components
The live model combines a structural-risk score (v4 = isotonic(kio_hist_5y + 2 × kio_recent_7d)) with the OONI-trajectory forecast (honest_forecast_v1); the served version is shown above. The logit-space blend (sigmoid(0.6·logit(v4)+0.4·logit(hf))) is auto-gated — it serves only on days it durably beats the live model, leave-one-country-out validated.
This score ranks which countries are heading into danger (forward country-ranking AUC 0.89 against Access Now KeepItOn labels, measured in the committed model-stats snapshot). It is not a within-country timing signal: in a strict past→future holdout, within-country timing AUC is 0.45, below chance. Live numbers: /v1/shutdown-risk/info.
- ensemble_raw
- 0
- hf_prob
- 25.5%
- sr_prob
- 0.1%
Honest validation
Same validation block this endpoint returns inline — never computed at request time, never adjusted per-country.
Model honesty caveats (5)
- Ensemble of two honestly-validated models. shutdown_risk_v4 provides country structural risk (the cross-country signal); honest_forecast_v1 provides OONI z-score trajectory (the within-country timing signal). The product captures WHICH and WHEN.
- Within-country median AUC 0.68. The model differentiates shutdown-precursor days from other days within the major shutdown-prone countries (MM, IN, RU, PK, IQ all >=0.70). For low-N countries the within-country signal is weaker — sample-size limited, not model-limited.
- Country structural risk dominates the cross-country signal; OONI measurement trajectory adds the within-country timing.
- City-level / sub-national shutdowns (~half of KeepItOn events) remain harder to predict from country-level aggregates.
- v1/v2/v3 lost the signal by over-fitting; v4 captured country-identity; v5 is the minimum ensemble that adds within-country timing on top of v4.
About this view
7-day internet-shutdown probability from shutdown_risk_v5. Trained and validated on Access Now KeepItOn STOP shutdowns (journalist-verified). Live JSON: /v1/shutdown-risk/FR.