Voidly
LB 🇱🇧 / country record

Lebanon’s archived Sentinel row dates and April 24 outage — a historical case

A published Sentinel row labeled 2026-04-17 recorded Lebanon above a 0.05 alert threshold; IODA recorded a North Lebanon BGP connectivity drop on 2026-04-24. This is a historical comparison of attributed row dates and incident evidence. Original forecast issuance and model lineage are unverified, so the interval does not establish advance warning or forecasting skill.

#sentinel#forecast#shutdown#lebanon#ioda#early-warning
Archived threshold crossing17 Apr 2026
Incident observation24 Apr 2026
Elapsed time between timestamps7.43 daysElapsed time between attributed archived row and IODA observation timestamps; original issuance and model lineage are unverified.
Incident IDLB-2026-0098 ↗
Historical 7-day forecastLebanon / 2026
0%10%20%2026-04-17: forecast 0.075, threshold 0.052026-04-18: forecast 0.069, threshold 0.052026-04-19: forecast 0.064, threshold 0.052026-04-20: forecast 0.071, threshold 0.052026-04-21: forecast 0.068, threshold 0.052026-04-22: forecast 0.07, threshold 0.052026-04-23: forecast 0.069, threshold 0.052026-04-24: forecast 0.077, threshold 0.052026-04-25: forecast 0.073, threshold 0.052026-04-26: forecast 0.081, threshold 0.052026-04-27: forecast 0.088, threshold 0.052026-04-28: forecast 0.08, threshold 0.052026-04-29: forecast 0.069, threshold 0.052026-04-30: forecast 0.099, threshold 0.052026-05-01: forecast 0.095, threshold 0.052026-05-02: forecast 0.087, threshold 0.052026-05-03: forecast 0.086, threshold 0.052026-05-04: forecast 0.1, threshold 0.052026-05-05: forecast 0.118, threshold 0.05Incident date2026-04-172026-04-262026-05-05
Archived forecast rowsThreshold in the published rows

Bundled case-study data, published 2026-05-21. This trajectory is historical; it does not show a current forecast or prove shutdown-onset skill.

Exact daily values & chart source

The published case attributes these rows to sentinel.db.sentinel_forecasts, labeled 02:00 UTC. Original issuance timestamps are not independently preserved. Values below come from the published case; this page makes no trajectory request. The incident marker uses its calendar date; the exact incident time is 2026-04-24T12:25:00Z.

Complete published trajectory
DateForecast probabilityAlert threshold
2026-04-170.0750.05
2026-04-180.0690.05
2026-04-190.0640.05
2026-04-200.0710.05
2026-04-210.0680.05
2026-04-220.070.05
2026-04-230.0690.05
2026-04-240.0770.05
2026-04-250.0730.05
2026-04-260.0810.05
2026-04-270.0880.05
2026-04-280.080.05
2026-04-290.0690.05
2026-04-300.0990.05
2026-05-010.0950.05
2026-05-020.0870.05
2026-05-030.0860.05
2026-05-040.10.05
2026-05-050.1180.05
Cite this case study

Use the permanent published URL and retain the case’s limitations.

Citation help

Published 2026-05-21

The complete account.

The account below describes historical rows and their source limits. Original issuance and model lineage remain unverified.

In this case published on 2026-05-21, the Sentinel row labeled 2026-04-17 at 02:00 UTC listed a next-7-day shutdown score of 0.075 against a threshold of 0.05. The published row sequence stayed above that threshold through May 5, when it reached 0.118. These are historical values attributed to Sentinel. Original issue-time and model-version receipts have not been independently preserved, so the dated rows do not establish when a reader could actually have received the scores.

On 2026-04-24 at 12:25 UTC, IODA recorded a critical BGP connectivity drop in North Lebanon. Voidly later paired the event with incident LB-2026-0098. The interval from the first published threshold-crossing timestamp to the incident timestamp is 7.43 days. That is a calculation between attributed dates, not verified forecast lead time.

What this case shows

The published rows, threshold, IODA observation and incident ID make the historical comparison inspectable. They do not establish that the scores were issued before the outcomes, or that the model can time Lebanon outages. One favorable alignment cannot measure skill against the local outage base rate.

The evaluation caveat

The original May 2026 account quoted 30-day precision and recall figures for Lebanon and the watched-country panel. Those historical calculations are not a qualified live scorecard: original issue times, model lineage, matching targets and complete mature outcomes have not been reconciled. We withhold those figures. See scorecard status and calibration status for the evaluation limits. This case remains an illustration of an archived alert and a citable incident, not evidence of prospective forecasting performance.

Historical model drivers

What moved the score.

SHAP contributions from this published run. Positive raises the score; negative lowers it. A feature contribution is not a causal explanation of the outage.

  • recent_shutdown+0.0900

    Historical 30-day recent-shutdown feature. The May 2026 account identified Lebanon’s earlier BGP outage cluster as a major model input.

  • block_rate_roll14_mean−0.0714

    Historical 14-day country measurement block-rate feature. The published SHAP contribution was -0.0714; it does not establish the cause of the later outage.

  • block_rate_roll7_mean−0.0607

    Historical 7-day country measurement block-rate feature. The published SHAP contribution was -0.0607; this is a model attribution, not independent evidence of what caused the outage.

These bundled feature values are historical. The forecast endpoint provides separate current output; its top_features field may differ from this case.

Follow the sources

The incident & the record.

LB-2026-0098 ↗

IODA detected a critical-level BGP outage on the North Lebanon regional entity (value 79 vs. history 83) on 2026-04-24 at 12:25 UTC. Published as incident LB-2026-0098.

Inspect incident JSON ↗

Sources cited by the study

External measurement evidence and Voidly’s own incident/evaluation links have different roles. Listing them together does not make each an independent confirmation.

This case is published under CC BY 4.0 in its original article metadata. Preserve attribution; upstream source material retains its own terms.