Static page over recorded historical data — not live. Every panel below was produced by python -m outage_radar demo from bundled recordings of real IODA measurements for past events (Sudan 2024, Iraq exam shutdowns, Iran 2022 and controls). It is a test of whether the detector gets known cases right, not a view of the internet now. Generated 2026-09-02 23:07 UTC; the underlying measurements are from the event windows named in each panel.

Data: IODA (Internet Outage Detection and Analysis), Georgia Tech. IODA data is Copyright © 2021–2025 Georgia Tech Research Corporation, All Rights Reserved — there is no Creative Commons grant. This page is non-commercial research and portfolio use only, and nothing here may be redistributed commercially without Georgia Tech's separate written permission. Signals used: BGP routed /24s (RouteViews and RIPE RIS), responsive /24s from active probing, and Merit network-telescope traffic. Google Transparency Report figures arrive through IODA under IODA's terms.

These are inferences from third-party traffic measurements, not observations of events. The tool cannot see intent — it sees packets stopping. “Deliberate shutdown” is a hypothesis with a stated weight of evidence, not a finding of fact about any government or operator. Corroborate against reporting on the ground before repeating it. Not an emergency service.
← index  ·  the frozen live snapshot →

outage-radar

Seeing a national internet outage in traffic measurements, and being careful about what that does and does not prove. Every panel below was produced from bundled recordings of real IODA measurements; the dashed line is the fitted seasonal baseline, and the shaded left-hand region is the window used to fit it.

Read this before quoting anything below. These are inferences from third-party traffic measurements, not observations of events. This tool cannot see intent: it sees packets stopping. A classification of “deliberate shutdown” is a hypothesis with a stated weight of evidence, not a finding of fact about any government, operator or incident, and it should be corroborated against reporting on the ground before it is repeated. It is not an emergency service and must not be relied on for safety-critical decisions. Where the evidence is thin, the tool says so; please carry that uncertainty forward rather than dropping it.

Sudan, February 2024 - nationwide connectivity collapse

2024-01-10 00:00 UTC → 2024-02-14 00:00 UTC

Sudan lost near-total national connectivity in early February 2024 during the ongoing conflict, with the major operators going offline. Attribution was contested at the time; the measurement record shows the collapse but not its cause, which is exactly the distinction this tool is trying to hold on to.

What a correct detector should say: SD: near-total loss on both the control plane (BGP) and the data plane, sustained for days, across every monitored operator. DE over the identical window is the control and should stay CLEAR.

IT'S HERE

Sudan SD

BGP routed /24s
2024-01-10 shaded left = baseline training window · dashed = expected · block = detected episode 2024-02-14
onset (local)
2024-02-02 16:30 (UTC+02:00)
onset shape
cliff — half the depth lost in 10 min
recovery
none within the window
depth
98% below seasonal baseline (z = -195.3)
duration
273.3 h — ongoing at window end
steps
3 distinct step-downs (staged escalation)
breadth
83% of 6 monitored networks affected
planes
BGP control plane 98% · reachability 85%
corroboration
corroborated by merit-nt, ping-slash24 - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)
schedule
fewer than 3 episodes to compare

Deliberate shutdown or ordered restriction possible

Deliberate shutdown or ordered restriction38%
Power failure, natural disaster or conflict damage26%
Submarine cable or upstream transit fault23%
Single-operator fault6%
Measurement artefact, not a real outage6%

Most consistent explanation: deliberate shutdown or ordered restriction (possible, 38% of the weighted evidence). Traffic fell off a cliff, shedding half its eventual depth within 30 minutes starting 16:30 local time, reaching 98% below the seasonal baseline, and lasting 273.3 hours. 83% of the 6 monitored networks in this country were affected at the same time. BGP routing fell 98% against a 85% fall in reachability. The main alternative not excluded by this evidence is power failure, natural disaster or conflict damage (26%). Held at 'possible' rather than 'likely': onset is not clock-aligned and does not repeat on a schedule. Asserting that a state ordered a shutdown requires more than this. This is an inference from traffic measurement alone. It is not a finding of fact about any government, operator or event, and it should be corroborated against reporting on the ground before being repeated.

Evidence (13 rules fired)
wtrulesupportsreasoning
-3.0BREADTH_NOT_SINGLESingle-operator faultRules out a single-operator fault.
-2.0CORROBORATEDMeasurement artefact, not a real outageTwo independent measurement methods agree (tier B), which makes a single-platform artefact unlikely.
-1.8PLANE_BOTH_NOT_CABLESubmarine cable or upstream transit faultArgues against a cable fault: a cut cable leaves the routing table largely intact.
+1.6LONG_DURATIONSubmarine cable or upstream transit faultDegraded for 11.4 days. Restoring a submarine cable needs a repair ship on station, which is a job of days to weeks, and the long tail is characteristic.
-1.6LONG_UNSCHEDULEDDeliberate shutdown or ordered restrictionArgues against an ordered restriction: it ran for days, began at no particular local time and never repeated on a schedule. Ordered shutdowns are normally bounded and tidy, because someone has to decide to end them.
+1.5SHAPE_CLIFFDeliberate shutdown or ordered restrictionOnset is a cliff: half the eventual loss happened within 30 minutes, consistent with a configuration change - though not with that alone.
+1.5SHAPE_CLIFF_CABLESubmarine cable or upstream transit faultA cable parting is equally instantaneous, so a cliff onset carries no weight either way between a cut and an order.
-1.5SHAPE_CLIFF_NOT_GRIDPower failure, natural disaster or conflict damageArgues against a grid or disaster cause: batteries and generators normally produce a decline over hours, not a step.
+1.4BREADTH_TOTALDeliberate shutdown or ordered restriction83% of monitored networks dropped simultaneously. Independent commercial operators do not fail together by chance; either they share infrastructure or they received the same instruction.
+1.2PLANE_BOTHDeliberate shutdown or ordered restrictionBoth planes collapsed together (BGP 98%, reachability 85%): networks were withdrawn from the routing table, so the outage is at the operator rather than in a cable.
+1.0PLANE_BOTH_POWERPower failure, natural disaster or conflict damageA total loss of power at core sites would also withdraw routes, so an infrastructure cause remains consistent with this.
+1.0BREADTH_TOTAL_POWERPower failure, natural disaster or conflict damageEqually consistent with a national grid failure hitting every operator at once.
+1.0LONG_DURATION_CONFLICTPower failure, natural disaster or conflict damageAlso consistent with sustained infrastructure damage or fuel exhaustion.

Alert level reasoning: sustained 98% drop below seasonal baseline for 273.3 h; deep national-scale drop; corroborated by merit-nt, ping-slash24 - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)

CLEAR

Germany DE

No sustained drop above the 20% watch threshold in the scoring window.

Iraq, June-July 2023 - recurring exam-period shutdowns

2023-06-01 00:00 UTC → 2023-07-10 00:00 UTC

Iraq has for several years imposed nationwide internet shutdowns during national exam sittings, in windows of a few hours on exam mornings. This is the least ambiguous class of deliberate shutdown because it is announced, bounded and repeated.

What a correct detector should say: IQ: repeated short outages beginning at essentially the same local clock time on many mornings, with cliff onset and cliff recovery. The repetition is the point: it is the strongest available evidence of a human schedule. GB is the control.

IT'S HERE

Iraq IQ

Responsive /24s (active probing)
2023-06-01 shaded left = baseline training window · dashed = expected · block = detected episode 2023-07-10
onset (local)
2023-06-26 04:00 (UTC+03:00)
onset shape
cliff — half the depth lost in 10 min
recovery
cliff (30 min)
depth
65% below seasonal baseline (z = -34.3)
duration
5.3 h
breadth
67% of 6 monitored networks affected
planes
BGP control plane 31% · reachability 65%
corroboration
corroborated by bgp - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)
schedule
7 onsets, interquartile spread 0 min of local time-of-day - repeats on a clock

Deliberate shutdown or ordered restriction likely

Deliberate shutdown or ordered restriction80%
Power failure, natural disaster or conflict damage5%
Submarine cable or upstream transit fault5%
Single-operator fault5%
Measurement artefact, not a real outage5%

Most consistent explanation: deliberate shutdown or ordered restriction (likely, 80% of the weighted evidence). Traffic fell off a cliff, shedding half its eventual depth within 30 minutes starting 04:00 local time, reaching 65% below the seasonal baseline, and lasting 5.3 hours. 67% of the 6 monitored networks in this country were affected at the same time. BGP routing fell 31% against a 65% fall in reachability. This is an inference from traffic measurement alone. It is not a finding of fact about any government, operator or event, and it should be corroborated against reporting on the ground before being repeated.

Evidence (10 rules fired)
wtrulesupportsreasoning
+2.6SCHEDULE_REPEATSDeliberate shutdown or ordered restrictionThe outage repeated 7 times at essentially the same local clock time. Repetition on a schedule is the strongest available evidence of a human decision; exam-period shutdowns look exactly like this.
-2.0SCHEDULE_NOT_PHYSICALPower failure, natural disaster or conflict damageArgues strongly against a physical cause: neither grids nor weather repeat to a timetable.
-2.0SCHEDULE_NOT_CABLESubmarine cable or upstream transit faultArgues strongly against a cable fault, which is a single event followed by a multi-day repair.
-2.0CORROBORATEDMeasurement artefact, not a real outageTwo independent measurement methods agree (tier B), which makes a single-platform artefact unlikely.
+1.5SHAPE_CLIFFDeliberate shutdown or ordered restrictionOnset is a cliff: half the eventual loss happened within 30 minutes, consistent with a configuration change - though not with that alone.
+1.5SHAPE_CLIFF_CABLESubmarine cable or upstream transit faultA cable parting is equally instantaneous, so a cliff onset carries no weight either way between a cut and an order.
-1.5SHAPE_CLIFF_NOT_GRIDPower failure, natural disaster or conflict damageArgues against a grid or disaster cause: batteries and generators normally produce a decline over hours, not a step.
+1.4RECOVERY_CLIFFDeliberate shutdown or ordered restrictionRecovery was as sharp as the onset: service came back in one step, as it does when a restriction is lifted rather than when equipment is repaired.
+1.2SHORT_WINDOWDeliberate shutdown or ordered restrictionA 5.3 hour window starting on a round local hour looks like a bounded restriction. Failures have no reason to end tidily.
+1.0CLOCK_ALIGNEDDeliberate shutdown or ordered restrictionOnset at 04:00 local time, within 10 minutes of a whole hour. Physical failures do not consult a wall clock - but a random onset lands this close to an hour one time in three, so on its own this is weak.

Alert level reasoning: sustained 65% drop below seasonal baseline for 5.3 h; deep national-scale drop; corroborated by bgp - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)

CLEAR

United Kingdom GB

No sustained drop above the 20% watch threshold in the scoring window.

West Africa, 14 March 2024 - submarine cable cuts

2024-02-05 00:00 UTC → 2024-03-28 00:00 UTC

On 14 March 2024 several submarine cable systems serving the West African coast (WACS, ACE, SAT-3 and MainOne) were damaged in the same period, cutting a large share of the region's international capacity. Restoration required cable ships and took weeks. This is the cleanest available example of an outage that is unambiguously physical and unambiguously not anyone's decision.

What a correct detector should say: Several neighbouring coastal countries degrade within the same few hours, at no particular local clock time, with a recovery measured in days rather than minutes. Cote d'Ivoire should be the deepest. Senegal sits north of the damage on different capacity and is the regional control: it should stay CLEAR, which is what stops 'the whole region dipped' from being read as 'every West African government acted at once'.

IT'S HERE

Cote D Ivoire CI

Responsive /24s (active probing)
2024-02-05 shaded left = baseline training window · dashed = expected · block = detected episode 2024-03-28
onset (local)
2024-03-13 09:45 (UTC+00:00)
onset shape
cliff — half the depth lost in 30 min
recovery
step (1.5 h)
depth
99% below seasonal baseline (z = -13.7)
duration
68.2 h
steps
5 distinct step-downs (staged escalation)
breadth
83% of 6 monitored networks affected
planes
BGP control plane 86% · reachability 99%
corroboration
corroborated by bgp - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)
schedule
3 onsets, interquartile spread 240 min of local time-of-day - no consistent clock time
regional
also dipping: GH, LR, NG

Submarine cable or upstream transit fault likely

Submarine cable or upstream transit fault58%
Power failure, natural disaster or conflict damage22%
Deliberate shutdown or ordered restriction10%
Single-operator fault5%
Measurement artefact, not a real outage5%

Most consistent explanation: submarine cable or upstream transit fault (likely, 58% of the weighted evidence). Traffic fell off a cliff, shedding half its eventual depth within 30 minutes starting 09:45 local time, reaching 99% below the seasonal baseline, and lasting 68.2 hours. 83% of the 6 monitored networks in this country were affected at the same time. BGP routing fell 86% against a 99% fall in reachability. 3 neighbouring monitored countries dipped in the same window. The main alternative not excluded by this evidence is power failure, natural disaster or conflict damage (22%). This is an inference from traffic measurement alone. It is not a finding of fact about any government, operator or event, and it should be corroborated against reporting on the ground before being repeated.

Evidence (16 rules fired)
wtrulesupportsreasoning
+3.5REGIONAL_COOCCURRENCESubmarine cable or upstream transit fault3 other monitored countries show an overlapping episode. Degradation that crosses borders points at infrastructure those countries share - a cable system or a common upstream - rather than at any one government. It takes a great deal more evidence to claim that several states acted in concert than to conclude that they sit on the same cable.
-3.0BREADTH_NOT_SINGLESingle-operator faultRules out a single-operator fault.
-2.1REGIONAL_NOT_ORDERDeliberate shutdown or ordered restrictionArgues against a single national order: 3 neighbouring countries fell with it.
-2.0CORROBORATEDMeasurement artefact, not a real outageTwo independent measurement methods agree (tier B), which makes a single-platform artefact unlikely.
-1.8PLANE_BOTH_NOT_CABLESubmarine cable or upstream transit faultArgues against a cable fault: a cut cable leaves the routing table largely intact.
+1.6LONG_DURATIONSubmarine cable or upstream transit faultDegraded for 2.8 days. Restoring a submarine cable needs a repair ship on station, which is a job of days to weeks, and the long tail is characteristic.
-1.6LONG_UNSCHEDULEDDeliberate shutdown or ordered restrictionArgues against an ordered restriction: it ran for days, began at no particular local time and never repeated on a schedule. Ordered shutdowns are normally bounded and tidy, because someone has to decide to end them.
+1.5SHAPE_CLIFFDeliberate shutdown or ordered restrictionOnset is a cliff: half the eventual loss happened within 30 minutes, consistent with a configuration change - though not with that alone.
+1.5SHAPE_CLIFF_CABLESubmarine cable or upstream transit faultA cable parting is equally instantaneous, so a cliff onset carries no weight either way between a cut and an order.
-1.5SHAPE_CLIFF_NOT_GRIDPower failure, natural disaster or conflict damageArgues against a grid or disaster cause: batteries and generators normally produce a decline over hours, not a step.
+1.4BREADTH_TOTALDeliberate shutdown or ordered restriction83% of monitored networks dropped simultaneously. Independent commercial operators do not fail together by chance; either they share infrastructure or they received the same instruction.
+1.2PLANE_BOTHDeliberate shutdown or ordered restrictionBoth planes collapsed together (BGP 86%, reachability 99%): networks were withdrawn from the routing table, so the outage is at the operator rather than in a cable.
+1.0PLANE_BOTH_POWERPower failure, natural disaster or conflict damageA total loss of power at core sites would also withdraw routes, so an infrastructure cause remains consistent with this.
+1.0BREADTH_TOTAL_POWERPower failure, natural disaster or conflict damageEqually consistent with a national grid failure hitting every operator at once.
+1.0LONG_DURATION_CONFLICTPower failure, natural disaster or conflict damageAlso consistent with sustained infrastructure damage or fuel exhaustion.
·CLOCK_UNTESTABLEMeasurement artefact, not a real outageThe sampling interval is too coarse to say whether the onset was aligned to a whole local hour, so the clock-alignment evidence is unavailable in both directions.

Alert level reasoning: sustained 99% drop below seasonal baseline for 68.2 h; deep national-scale drop; corroborated by bgp - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)

IT'S HERE

Ghana GH

BGP routed /24s
2024-02-05 shaded left = baseline training window · dashed = expected · block = detected episode 2024-03-28
onset (local)
2024-03-14 10:30 (UTC+00:00)
onset shape
cliff — half the depth lost in 15 min
recovery
none within the window
depth
85% below seasonal baseline (z = -169.5)
duration
325.2 h — ongoing at window end
steps
2 distinct step-downs (staged escalation)
breadth
83% of 6 monitored networks affected
planes
BGP control plane 85% · reachability 82%
corroboration
corroborated by ping-slash24 - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)
schedule
fewer than 3 episodes to compare
regional
also dipping: CI, LR, NG, SN

Submarine cable or upstream transit fault likely

Submarine cable or upstream transit fault63%
Power failure, natural disaster or conflict damage21%
Deliberate shutdown or ordered restriction5%
Single-operator fault5%
Measurement artefact, not a real outage5%

Most consistent explanation: submarine cable or upstream transit fault (likely, 63% of the weighted evidence). Traffic fell off a cliff, shedding half its eventual depth within 30 minutes starting 10:30 local time, reaching 85% below the seasonal baseline, and lasting 325.2 hours. 83% of the 6 monitored networks in this country were affected at the same time. BGP routing fell 85% against a 82% fall in reachability. 4 neighbouring monitored countries dipped in the same window. The main alternative not excluded by this evidence is power failure, natural disaster or conflict damage (21%). This is an inference from traffic measurement alone. It is not a finding of fact about any government, operator or event, and it should be corroborated against reporting on the ground before being repeated.

Evidence (16 rules fired)
wtrulesupportsreasoning
+4.2REGIONAL_COOCCURRENCESubmarine cable or upstream transit fault4 other monitored countries show an overlapping episode. Degradation that crosses borders points at infrastructure those countries share - a cable system or a common upstream - rather than at any one government. It takes a great deal more evidence to claim that several states acted in concert than to conclude that they sit on the same cable.
-3.0BREADTH_NOT_SINGLESingle-operator faultRules out a single-operator fault.
-2.8REGIONAL_NOT_ORDERDeliberate shutdown or ordered restrictionArgues against a single national order: 4 neighbouring countries fell with it.
-2.0CORROBORATEDMeasurement artefact, not a real outageTwo independent measurement methods agree (tier B), which makes a single-platform artefact unlikely.
-1.8PLANE_BOTH_NOT_CABLESubmarine cable or upstream transit faultArgues against a cable fault: a cut cable leaves the routing table largely intact.
+1.6LONG_DURATIONSubmarine cable or upstream transit faultDegraded for 13.6 days. Restoring a submarine cable needs a repair ship on station, which is a job of days to weeks, and the long tail is characteristic.
-1.6LONG_UNSCHEDULEDDeliberate shutdown or ordered restrictionArgues against an ordered restriction: it ran for days, began at no particular local time and never repeated on a schedule. Ordered shutdowns are normally bounded and tidy, because someone has to decide to end them.
+1.5SHAPE_CLIFFDeliberate shutdown or ordered restrictionOnset is a cliff: half the eventual loss happened within 30 minutes, consistent with a configuration change - though not with that alone.
+1.5SHAPE_CLIFF_CABLESubmarine cable or upstream transit faultA cable parting is equally instantaneous, so a cliff onset carries no weight either way between a cut and an order.
-1.5SHAPE_CLIFF_NOT_GRIDPower failure, natural disaster or conflict damageArgues against a grid or disaster cause: batteries and generators normally produce a decline over hours, not a step.
+1.4BREADTH_TOTALDeliberate shutdown or ordered restriction83% of monitored networks dropped simultaneously. Independent commercial operators do not fail together by chance; either they share infrastructure or they received the same instruction.
+1.2PLANE_BOTHDeliberate shutdown or ordered restrictionBoth planes collapsed together (BGP 85%, reachability 82%): networks were withdrawn from the routing table, so the outage is at the operator rather than in a cable.
+1.0PLANE_BOTH_POWERPower failure, natural disaster or conflict damageA total loss of power at core sites would also withdraw routes, so an infrastructure cause remains consistent with this.
+1.0BREADTH_TOTAL_POWERPower failure, natural disaster or conflict damageEqually consistent with a national grid failure hitting every operator at once.
+1.0LONG_DURATION_CONFLICTPower failure, natural disaster or conflict damageAlso consistent with sustained infrastructure damage or fuel exhaustion.
·CLOCK_UNTESTABLEMeasurement artefact, not a real outageThe sampling interval is too coarse to say whether the onset was aligned to a whole local hour, so the clock-alignment evidence is unavailable in both directions.

Alert level reasoning: sustained 85% drop below seasonal baseline for 325.2 h; deep national-scale drop; corroborated by ping-slash24 - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)

IT'S HERE

Liberia LR

Responsive /24s (active probing)
2024-02-05 shaded left = baseline training window · dashed = expected · block = detected episode 2024-03-28
onset (local)
2024-03-14 05:00 (UTC+00:00)
onset shape
cliff — half the depth lost in 15 min
recovery
ramp (7.0 h)
depth
81% below seasonal baseline (z = -23.0)
duration
33.5 h
breadth
50% of 6 monitored networks affected
planes
BGP control plane 63% · reachability 81%
corroboration
corroborated by bgp - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)
schedule
8 onsets, interquartile spread 390 min of local time-of-day - no consistent clock time
regional
also dipping: CI, GH, NG

Submarine cable or upstream transit fault likely

Submarine cable or upstream transit fault56%
Deliberate shutdown or ordered restriction17%
Power failure, natural disaster or conflict damage12%
Single-operator fault8%
Measurement artefact, not a real outage8%

Most consistent explanation: submarine cable or upstream transit fault (likely, 56% of the weighted evidence). Traffic fell off a cliff, shedding half its eventual depth within 30 minutes starting 05:00 local time, reaching 81% below the seasonal baseline, and lasting 33.5 hours. 50% of the 6 monitored networks in this country were affected at the same time. BGP routing fell 63% against a 81% fall in reachability. 3 neighbouring monitored countries dipped in the same window. The main alternative not excluded by this evidence is deliberate shutdown or ordered restriction (17%). This is an inference from traffic measurement alone. It is not a finding of fact about any government, operator or event, and it should be corroborated against reporting on the ground before being repeated.

Evidence (11 rules fired)
wtrulesupportsreasoning
+3.5REGIONAL_COOCCURRENCESubmarine cable or upstream transit fault3 other monitored countries show an overlapping episode. Degradation that crosses borders points at infrastructure those countries share - a cable system or a common upstream - rather than at any one government. It takes a great deal more evidence to claim that several states acted in concert than to conclude that they sit on the same cable.
-2.1REGIONAL_NOT_ORDERDeliberate shutdown or ordered restrictionArgues against a single national order: 3 neighbouring countries fell with it.
-2.0CORROBORATEDMeasurement artefact, not a real outageTwo independent measurement methods agree (tier B), which makes a single-platform artefact unlikely.
-1.8PLANE_BOTH_NOT_CABLESubmarine cable or upstream transit faultArgues against a cable fault: a cut cable leaves the routing table largely intact.
+1.5SHAPE_CLIFFDeliberate shutdown or ordered restrictionOnset is a cliff: half the eventual loss happened within 30 minutes, consistent with a configuration change - though not with that alone.
+1.5SHAPE_CLIFF_CABLESubmarine cable or upstream transit faultA cable parting is equally instantaneous, so a cliff onset carries no weight either way between a cut and an order.
-1.5SHAPE_CLIFF_NOT_GRIDPower failure, natural disaster or conflict damageArgues against a grid or disaster cause: batteries and generators normally produce a decline over hours, not a step.
+1.2PLANE_BOTHDeliberate shutdown or ordered restrictionBoth planes collapsed together (BGP 63%, reachability 81%): networks were withdrawn from the routing table, so the outage is at the operator rather than in a cable.
+1.0PLANE_BOTH_POWERPower failure, natural disaster or conflict damageA total loss of power at core sites would also withdraw routes, so an infrastructure cause remains consistent with this.
+0.8RECOVERY_RAMPPower failure, natural disaster or conflict damageRecovery was gradual over hours, consistent with crews restoring sites individually.
·CLOCK_UNTESTABLEMeasurement artefact, not a real outageThe sampling interval is too coarse to say whether the onset was aligned to a whole local hour, so the clock-alignment evidence is unavailable in both directions.

Alert level reasoning: sustained 81% drop below seasonal baseline for 33.5 h; deep national-scale drop; corroborated by bgp - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)

WARNING

Senegal SN

Responsive /24s (active probing)
2024-02-05 shaded left = baseline training window · dashed = expected · block = detected episode 2024-03-28
onset (local)
2024-03-18 05:15 (UTC+00:00)
onset shape
ramp — half the depth lost in 22.2 h
recovery
cliff (15 min)
depth
52% below seasonal baseline (z = -73.2)
duration
1.0 h
breadth
50% of 6 monitored networks affected
planes
BGP control plane 0% · reachability 52%
corroboration
no independent source shows an overlapping episode
schedule
fewer than 3 episodes to compare
regional
also dipping: GH

Submarine cable or upstream transit fault possible

Submarine cable or upstream transit fault48%
Power failure, natural disaster or conflict damage25%
Measurement artefact, not a real outage18%
Deliberate shutdown or ordered restriction5%
Single-operator fault5%

Most consistent explanation: submarine cable or upstream transit fault (possible, 48% of the weighted evidence). Traffic declined gradually, taking over 2 hours to lose half its depth starting 05:15 local time, reaching 52% below the seasonal baseline, and lasting 1.0 hours. 50% of the 6 monitored networks in this country were affected at the same time. BGP routing fell 0% against a 52% fall in reachability. 1 neighbouring monitored countries dipped in the same window. The main alternative not excluded by this evidence is power failure, natural disaster or conflict damage (25%). This is an inference from traffic measurement alone. It is not a finding of fact about any government, operator or event, and it should be corroborated against reporting on the ground before being repeated.

Evidence (8 rules fired)
wtrulesupportsreasoning
+2.6PLANE_DATA_ONLYSubmarine cable or upstream transit faultReachability fell 52% while BGP announcements only fell 0%: the prefixes are still advertised but packets are not getting through. That is the classic fingerprint of a severed cable or a failed upstream transit path, not of a network being switched off.
+2.2SHAPE_RAMPPower failure, natural disaster or conflict damageOnset is a ramp over two hours or more, consistent with sites failing one by one as battery and generator reserves are exhausted.
+2.1REGIONAL_COOCCURRENCESubmarine cable or upstream transit fault1 other monitored countries show an overlapping episode. Degradation that crosses borders points at infrastructure those countries share - a cable system or a common upstream - rather than at any one government. It takes a great deal more evidence to claim that several states acted in concert than to conclude that they sit on the same cable.
+1.5UNCORROBORATEDMeasurement artefact, not a real outageNo second measurement method shows the same event, so a fault in one platform's collection cannot be excluded.
-1.2SHAPE_RAMP_NOT_ORDERDeliberate shutdown or ordered restrictionArgues against an ordered shutdown, which is normally executed in minutes.
-0.8PLANE_DATA_ONLY_NOT_ORDERDeliberate shutdown or ordered restrictionArgues somewhat against an ordered shutdown, which more often withdraws routes as well - though filtering at the edge can also spare the control plane.
-0.7REGIONAL_NOT_ORDERDeliberate shutdown or ordered restrictionArgues against a single national order: 1 neighbouring countries fell with it.
·CLOCK_UNTESTABLEMeasurement artefact, not a real outageThe sampling interval is too coarse to say whether the onset was aligned to a whole local hour, so the clock-alignment evidence is unavailable in both directions.

Alert level reasoning: sustained 52% drop below seasonal baseline for 1.0 h; deep national-scale drop; held at WARNING: no independent second source agrees yet

WATCH

Nigeria NG

Responsive /24s (active probing)
2024-02-05 shaded left = baseline training window · dashed = expected · block = detected episode 2024-03-28
onset (local)
2024-03-14 11:30 (UTC+01:00)
onset shape
cliff — half the depth lost in 30 min
recovery
ramp (2.0 h)
depth
30% below seasonal baseline (z = -29.1)
duration
15.2 h
steps
2 distinct step-downs (staged escalation)
breadth
50% of 6 monitored networks affected
planes
BGP control plane 15% · reachability 30%
corroboration
corroborated by merit-nt - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)
schedule
3 onsets, interquartile spread 270 min of local time-of-day - no consistent clock time
regional
also dipping: CI, GH, LR

Submarine cable or upstream transit fault likely

Submarine cable or upstream transit fault73%
Deliberate shutdown or ordered restriction7%
Power failure, natural disaster or conflict damage7%
Single-operator fault7%
Measurement artefact, not a real outage7%

Most consistent explanation: submarine cable or upstream transit fault (likely, 73% of the weighted evidence). Traffic fell off a cliff, shedding half its eventual depth within 30 minutes starting 11:30 local time, reaching 30% below the seasonal baseline, and lasting 15.2 hours. 50% of the 6 monitored networks in this country were affected at the same time. BGP routing fell 15% against a 30% fall in reachability. 3 neighbouring monitored countries dipped in the same window. This is an inference from traffic measurement alone. It is not a finding of fact about any government, operator or event, and it should be corroborated against reporting on the ground before being repeated.

Evidence (8 rules fired)
wtrulesupportsreasoning
+3.5REGIONAL_COOCCURRENCESubmarine cable or upstream transit fault3 other monitored countries show an overlapping episode. Degradation that crosses borders points at infrastructure those countries share - a cable system or a common upstream - rather than at any one government. It takes a great deal more evidence to claim that several states acted in concert than to conclude that they sit on the same cable.
-2.1REGIONAL_NOT_ORDERDeliberate shutdown or ordered restrictionArgues against a single national order: 3 neighbouring countries fell with it.
-2.0CORROBORATEDMeasurement artefact, not a real outageTwo independent measurement methods agree (tier B), which makes a single-platform artefact unlikely.
+1.5SHAPE_CLIFFDeliberate shutdown or ordered restrictionOnset is a cliff: half the eventual loss happened within 30 minutes, consistent with a configuration change - though not with that alone.
+1.5SHAPE_CLIFF_CABLESubmarine cable or upstream transit faultA cable parting is equally instantaneous, so a cliff onset carries no weight either way between a cut and an order.
-1.5SHAPE_CLIFF_NOT_GRIDPower failure, natural disaster or conflict damageArgues against a grid or disaster cause: batteries and generators normally produce a decline over hours, not a step.
+0.8RECOVERY_RAMPPower failure, natural disaster or conflict damageRecovery was gradual over hours, consistent with crews restoring sites individually.
·CLOCK_UNTESTABLEMeasurement artefact, not a real outageThe sampling interval is too coarse to say whether the onset was aligned to a whole local hour, so the clock-alignment evidence is unavailable in both directions.

Alert level reasoning: sustained 30% drop below seasonal baseline for 15.2 h; corroborated by merit-nt - different measurement method but the same aggregation platform (tier B: independent sensors, shared ingestion)

East Africa, 18 March 2024 - a synchronised dip that is probably NOT an outage

2024-02-15 00:00 UTC → 2024-03-25 00:00 UTC

While building the fixtures for the March 2024 East African cable damage, the country-level IODA aggregates turned out not to show it: daily medians for Tanzania and Kenya are flat across the whole window. What the detector did find was an hour-long simultaneous dip on 18 March in all three countries at the same UTC minute, which is far more consistent with a gap in the active prober than with anything happening in East Africa. It is kept as a fixture deliberately, because a tool like this is only trustworthy if it is willing to say 'this is probably my own instrument'.

What a correct detector should say: A short dip appears in three countries beginning at the identical UTC sample and recovering within about an hour. A detector that reports this as a regional shutdown is wrong. The correct answer is to flag it as a probable measurement artefact: real events do not start at the same instant in three countries and then stop, and the countries' daily medians are unchanged throughout. This case exists to prove the artefact guard fires.

WATCH

Tanzania TZ

Responsive /24s (active probing)
2024-02-15 shaded left = baseline training window · dashed = expected · block = detected episode 2024-03-25
onset (local)
2024-03-18 08:20 (UTC+03:00)
onset shape
cliff — half the depth lost in 10 min
recovery
cliff (10 min)
depth
54% below seasonal baseline (z = -44.7)
duration
1.0 h
breadth
100% of 6 monitored networks affected
planes
BGP control plane 1% · reachability 54%
corroboration
no independent source shows an overlapping episode
schedule
fewer than 3 episodes to compare
regional
also dipping: KE, UG

Measurement artefact, not a real outage likely

Measurement artefact, not a real outage57%
Submarine cable or upstream transit fault25%
Deliberate shutdown or ordered restriction9%
Power failure, natural disaster or conflict damage5%
Single-operator fault5%

Most consistent explanation: measurement artefact, not a real outage (likely, 57% of the weighted evidence). Traffic fell off a cliff, shedding half its eventual depth within 30 minutes starting 08:20 local time, reaching 54% below the seasonal baseline, and lasting 1.0 hours. 100% of the 6 monitored networks in this country were affected at the same time. BGP routing fell 1% against a 54% fall in reachability. 2 neighbouring monitored countries dipped in the same window. The main alternative not excluded by this evidence is submarine cable or upstream transit fault (25%). This is an inference from traffic measurement alone. It is not a finding of fact about any government, operator or event, and it should be corroborated against reporting on the ground before being repeated.

Evidence (13 rules fired)
wtrulesupportsreasoning
+4.0SYNCHRONISED_ARTEFACTMeasurement artefact, not a real outageThe dip begins in 2 other countries within the same measurement sample and clears within a few hours. Independent national outages do not start on the same tick; a gap in the measurement platform does. Treat this as an instrument fault until a second platform shows it.
-3.0BREADTH_NOT_SINGLESingle-operator faultRules out a single-operator fault.
-3.0SYNCHRONISED_NOT_REALDeliberate shutdown or ordered restrictionArgues strongly against any real-world cause, including an ordered shutdown.
+2.6PLANE_DATA_ONLYSubmarine cable or upstream transit faultReachability fell 54% while BGP announcements only fell 1%: the prefixes are still advertised but packets are not getting through. That is the classic fingerprint of a severed cable or a failed upstream transit path, not of a network being switched off.
-2.0SYNCHRONISED_NOT_CABLESubmarine cable or upstream transit faultArgues against a cable fault: a cut does not repair itself within the hour.
+1.5SHAPE_CLIFFDeliberate shutdown or ordered restrictionOnset is a cliff: half the eventual loss happened within 30 minutes, consistent with a configuration change - though not with that alone.
+1.5SHAPE_CLIFF_CABLESubmarine cable or upstream transit faultA cable parting is equally instantaneous, so a cliff onset carries no weight either way between a cut and an order.
+1.5UNCORROBORATEDMeasurement artefact, not a real outageNo second measurement method shows the same event, so a fault in one platform's collection cannot be excluded.
-1.5SHAPE_CLIFF_NOT_GRIDPower failure, natural disaster or conflict damageArgues against a grid or disaster cause: batteries and generators normally produce a decline over hours, not a step.
+1.4RECOVERY_CLIFFDeliberate shutdown or ordered restrictionRecovery was as sharp as the onset: service came back in one step, as it does when a restriction is lifted rather than when equipment is repaired.
+1.4BREADTH_TOTALDeliberate shutdown or ordered restriction100% of monitored networks dropped simultaneously. Independent commercial operators do not fail together by chance; either they share infrastructure or they received the same instruction.
+1.0BREADTH_TOTAL_POWERPower failure, natural disaster or conflict damageEqually consistent with a national grid failure hitting every operator at once.
-0.8PLANE_DATA_ONLY_NOT_ORDERDeliberate shutdown or ordered restrictionArgues somewhat against an ordered shutdown, which more often withdraws routes as well - though filtering at the edge can also spare the control plane.

Alert level reasoning: downgraded to WATCH: the leading explanation is a measurement artefact (57%), not an event on the ground - sustained 54% drop below seasonal baseline for 1.0 h; deep national-scale drop; held at WARNING: no independent second source agrees yet

WATCH

Uganda UG

Responsive /24s (active probing)
2024-02-15 shaded left = baseline training window · dashed = expected · block = detected episode 2024-03-25
onset (local)
2024-03-18 08:20 (UTC+03:00)
onset shape
cliff — half the depth lost in 10 min
recovery
cliff (10 min)
depth
55% below seasonal baseline (z = -32.0)
duration
1.0 h
breadth
33% of 6 monitored networks affected
planes
BGP control plane 0% · reachability 55%
corroboration
no independent source shows an overlapping episode
schedule
fewer than 3 episodes to compare
regional
also dipping: TZ, KE

Measurement artefact, not a real outage possible

Measurement artefact, not a real outage46%
Single-operator fault27%
Submarine cable or upstream transit fault20%
Deliberate shutdown or ordered restriction4%
Power failure, natural disaster or conflict damage4%

Most consistent explanation: measurement artefact, not a real outage (possible, 46% of the weighted evidence). Traffic fell off a cliff, shedding half its eventual depth within 30 minutes starting 08:20 local time, reaching 55% below the seasonal baseline, and lasting 1.0 hours. 33% of the 6 monitored networks in this country were affected at the same time. BGP routing fell 0% against a 55% fall in reachability. 2 neighbouring monitored countries dipped in the same window. The main alternative not excluded by this evidence is single-operator fault (27%). This is an inference from traffic measurement alone. It is not a finding of fact about any government, operator or event, and it should be corroborated against reporting on the ground before being repeated.

Evidence (13 rules fired)
wtrulesupportsreasoning
+4.0SYNCHRONISED_ARTEFACTMeasurement artefact, not a real outageThe dip begins in 2 other countries within the same measurement sample and clears within a few hours. Independent national outages do not start on the same tick; a gap in the measurement platform does. Treat this as an instrument fault until a second platform shows it.
+3.0BREADTH_NARROWSingle-operator faultOnly 33% of monitored networks are affected while the rest of the country is unchanged. This is an operator problem, not a national event.
-3.0SYNCHRONISED_NOT_REALDeliberate shutdown or ordered restrictionArgues strongly against any real-world cause, including an ordered shutdown.
+2.6PLANE_DATA_ONLYSubmarine cable or upstream transit faultReachability fell 55% while BGP announcements only fell 0%: the prefixes are still advertised but packets are not getting through. That is the classic fingerprint of a severed cable or a failed upstream transit path, not of a network being switched off.
-2.5BREADTH_NARROW_NOT_ORDERDeliberate shutdown or ordered restrictionArgues strongly against a national order, which would reach all licensees.
-2.0SYNCHRONISED_NOT_CABLESubmarine cable or upstream transit faultArgues against a cable fault: a cut does not repair itself within the hour.
+1.5SHAPE_CLIFFDeliberate shutdown or ordered restrictionOnset is a cliff: half the eventual loss happened within 30 minutes, consistent with a configuration change - though not with that alone.
+1.5SHAPE_CLIFF_CABLESubmarine cable or upstream transit faultA cable parting is equally instantaneous, so a cliff onset carries no weight either way between a cut and an order.
+1.5UNCORROBORATEDMeasurement artefact, not a real outageNo second measurement method shows the same event, so a fault in one platform's collection cannot be excluded.
-1.5SHAPE_CLIFF_NOT_GRIDPower failure, natural disaster or conflict damageArgues against a grid or disaster cause: batteries and generators normally produce a decline over hours, not a step.
-1.5BREADTH_NARROW_NOT_POWERPower failure, natural disaster or conflict damageArgues against a widespread physical cause.
+1.4RECOVERY_CLIFFDeliberate shutdown or ordered restrictionRecovery was as sharp as the onset: service came back in one step, as it does when a restriction is lifted rather than when equipment is repaired.
-0.8PLANE_DATA_ONLY_NOT_ORDERDeliberate shutdown or ordered restrictionArgues somewhat against an ordered shutdown, which more often withdraws routes as well - though filtering at the edge can also spare the control plane.

Alert level reasoning: sustained 55% drop below seasonal baseline for 1.0 h; deep drop but narrow: only 33% of monitored networks affected, so this looks like one operator rather than the country

WATCH

Kenya KE

Responsive /24s (active probing)
2024-02-15 shaded left = baseline training window · dashed = expected · block = detected episode 2024-03-25
onset (local)
2024-03-18 08:20 (UTC+03:00)
onset shape
cliff — half the depth lost in 10 min
recovery
cliff (10 min)
depth
48% below seasonal baseline (z = -49.1)
duration
1.0 h
breadth
83% of 6 monitored networks affected
planes
BGP control plane -0% · reachability 48%
corroboration
no independent source shows an overlapping episode
schedule
fewer than 3 episodes to compare
regional
also dipping: TZ, UG

Measurement artefact, not a real outage likely

Measurement artefact, not a real outage57%
Submarine cable or upstream transit fault25%
Deliberate shutdown or ordered restriction9%
Power failure, natural disaster or conflict damage5%
Single-operator fault5%

Most consistent explanation: measurement artefact, not a real outage (likely, 57% of the weighted evidence). Traffic fell off a cliff, shedding half its eventual depth within 30 minutes starting 08:20 local time, reaching 48% below the seasonal baseline, and lasting 1.0 hours. 83% of the 6 monitored networks in this country were affected at the same time. BGP routing fell -0% against a 48% fall in reachability. 2 neighbouring monitored countries dipped in the same window. The main alternative not excluded by this evidence is submarine cable or upstream transit fault (25%). This is an inference from traffic measurement alone. It is not a finding of fact about any government, operator or event, and it should be corroborated against reporting on the ground before being repeated.

Evidence (13 rules fired)
wtrulesupportsreasoning
+4.0SYNCHRONISED_ARTEFACTMeasurement artefact, not a real outageThe dip begins in 2 other countries within the same measurement sample and clears within a few hours. Independent national outages do not start on the same tick; a gap in the measurement platform does. Treat this as an instrument fault until a second platform shows it.
-3.0BREADTH_NOT_SINGLESingle-operator faultRules out a single-operator fault.
-3.0SYNCHRONISED_NOT_REALDeliberate shutdown or ordered restrictionArgues strongly against any real-world cause, including an ordered shutdown.
+2.6PLANE_DATA_ONLYSubmarine cable or upstream transit faultReachability fell 48% while BGP announcements only fell -0%: the prefixes are still advertised but packets are not getting through. That is the classic fingerprint of a severed cable or a failed upstream transit path, not of a network being switched off.
-2.0SYNCHRONISED_NOT_CABLESubmarine cable or upstream transit faultArgues against a cable fault: a cut does not repair itself within the hour.
+1.5SHAPE_CLIFFDeliberate shutdown or ordered restrictionOnset is a cliff: half the eventual loss happened within 30 minutes, consistent with a configuration change - though not with that alone.
+1.5SHAPE_CLIFF_CABLESubmarine cable or upstream transit faultA cable parting is equally instantaneous, so a cliff onset carries no weight either way between a cut and an order.
+1.5UNCORROBORATEDMeasurement artefact, not a real outageNo second measurement method shows the same event, so a fault in one platform's collection cannot be excluded.
-1.5SHAPE_CLIFF_NOT_GRIDPower failure, natural disaster or conflict damageArgues against a grid or disaster cause: batteries and generators normally produce a decline over hours, not a step.
+1.4RECOVERY_CLIFFDeliberate shutdown or ordered restrictionRecovery was as sharp as the onset: service came back in one step, as it does when a restriction is lifted rather than when equipment is repaired.
+1.4BREADTH_TOTALDeliberate shutdown or ordered restriction83% of monitored networks dropped simultaneously. Independent commercial operators do not fail together by chance; either they share infrastructure or they received the same instruction.
+1.0BREADTH_TOTAL_POWERPower failure, natural disaster or conflict damageEqually consistent with a national grid failure hitting every operator at once.
-0.8PLANE_DATA_ONLY_NOT_ORDERDeliberate shutdown or ordered restrictionArgues somewhat against an ordered shutdown, which more often withdraws routes as well - though filtering at the edge can also spare the control plane.

Alert level reasoning: downgraded to WATCH: the leading explanation is a measurement artefact (57%), not an event on the ground - sustained 48% drop below seasonal baseline for 1.0 h; sharp cliff-edge onset at national scale; held at WARNING: no independent second source agrees yet

Iran, September 2022 - protest-period restrictions

2022-08-27 00:00 UTC → 2022-10-01 00:00 UTC

During the protests that began in September 2022, mobile internet in Iran was repeatedly restricted. The restrictions hit mobile networks while fixed-line connectivity largely persisted, and country-level aggregates of routed and responsive /24s are dominated by fixed-line infrastructure, so the national signal barely moves. This is the single most important case in the set: it shows that a CLEAR from this tool means 'no large country-wide drop in these particular signals', not 'no shutdown'. Detecting mobile-only restrictions needs per-ASN monitoring of the mobile operators, or a measurement platform that sees client traffic rather than infrastructure.

What a correct detector should say: A KNOWN FALSE NEGATIVE, kept deliberately. The correct output is CLEAR with a sub-threshold note, not a WATCH. Country-level routed-prefix counts dip about 7.6% on the night of 21 September 2022 at z = -15 - statistically unmistakable, and nowhere near the 20% threshold this tool escalates on. Anyone reading the ladder alone would conclude nothing happened in Iran, which is wrong, and that is exactly the limitation a user needs to know about before trusting a CLEAR.

CLEAR

Iran (Islamic Republic Of) IR

No sustained drop above the 20% watch threshold in the scoring window.