CanaryObservatory
 
Infrastructure & cyber · Instrument dossier

Internet Connectivity

Country-level internet outage events detected by Georgia Tech's Internet Outage Detection and Analysis (IODA) project from BGP, active probing and telescope data. Sudden national connectivity drops can reflect government shutdowns, power failures, cable cuts or conflict.

Current reading

8 open country outage events at the latest collection. Not enough data yet for a historical comparison.

Reading time:

Global, country level. Regions are inferred from IODA entities and are not plotted on the map.

Collector availability
Live
Collection freshness
Last successful collection .

Historical comparison

Internet Connectivity at each collection · breaks = gaps (not interpolated) · expected range appears once 5 earlier days exist at each hour

Interpretation

Not enough data yet. The available history cannot establish whether this reading is typical or unusual.

What movement could mean
  • Government-ordered shutdowns
  • Power grid failures
  • Submarine cable damage

Not established. See the Connectivity Hypothesis experiment. Source-reported severity, statistical deviation and real-world risk are different questions. This reading does not independently support a hypothesis about intent or cause.

Evidence and limitations

Cadence and latency
Collection is scheduled every 30 minutes. Collection cadence is not source latency; no measured end-to-end latency is provided here. Source publication and revisions can delay what is visible.
Known false-positive patterns
  • Measurement infrastructure problems
  • Routing changes without user impact
Coverage and blind spots
  • IODA scores are relative to each country's normal signal; they are not a percentage.
  • Small countries and territories produce many events from minor changes.
  • Events can be revised or extended after detection.
Technical detail
Measurement definition
Each observation is one IODA outage event for a country and data source; the value is IODA's outage score. The baseline tracks how many country outage events are open at each collection.
Raw metric
open_country_outages
Method
The collector reads country outage events from the previous 7 days. Events are keyed by country, data source and start time. IODA scores are not comparable across countries of different sizes.
Stored observations
175

Normal versus unusual

A baseline cannot yet be established. The observatory needs more history before it can say what "normal" looks like for this signal, so no anomaly score is shown.

4 historical samples so far · Snapshot count vs. historical snapshots

Source severity

The label the publishing agency assigns (e.g. NWS 'Severe'). Shown verbatim.

Statistical anomaly

How far today's value sits from this signal's own history. Says nothing about consequences.

Real-world risk

A judgment the observatory does not make. Consult official sources for safety decisions.

Observations in range

Recent collection runs

  • · success · 162 stored
  • · success · 162 stored
  • · success · 163 stored
  • · success · 163 stored
  • · success · 164 stored
  • · failed · 0 stored · Interrupted before finishing (no result recorded)
  • · failed · 0 stored · Interrupted before finishing (no result recorded)
  • · failed · 0 stored · Interrupted before finishing (no result recorded)
  • · success · 164 stored
  • · success · 164 stored
  • · success · 164 stored
  • · success · 166 stored

How this works

We read country-level internet outage events from Georgia Tech's IODA project every 30 minutes. IODA spots outages by watching internet routing, probes and background traffic for each country.

The number we watch is how many countries have an open outage event. "Normal" is that count at the same hour of the day on earlier days (at least 5).

Limits: outages are country-wide, so they are not plotted on the map. IODA's scores are relative to each country, not percentages, and small countries can look dramatic.

FAQ

Is an outage a shutdown?
Not necessarily. Power cuts, cable damage and technical faults look the same in this data as deliberate shutdowns.
Why isn't it on the map?
A country-wide event has no meaningful point position, and drawing one would suggest a precision we don't have.