AI Agent - Aug 8, 2026

Cloudflare Radar Researcher Internet Outage Analysis Guide

Quick answer

Ask Radar Researcher a bounded outage question: name the country, region, ASN, or domain; use an absolute time window and timezone; specify the metric; and request a baseline comparison. Then inspect the Radar API endpoint, parameters, returned data, chart, and trace.

Report the result first as an observation from Cloudflare Radar. Only attribute cause after corroborating with routing evidence, operator status, official statements, local reporting, other measurement networks, and domain experts.

Start with a falsifiable question

Weak: “What happened to the Internet yesterday?”

Better: “For location XX, compare Radar’s HTTP traffic and recorded outage events from 2026-08-07 00:00 to 2026-08-08 00:00 UTC against the same weekdays over the prior four weeks. Identify the start, minimum, recovery, and data gaps. Do not infer cause.”

The stronger prompt defines scope, evidence, baseline, and the conclusion boundary.

Select metrics deliberately

QuestionCandidate Radar evidenceCaveat
Did HTTP activity fall?HTTP traffic time series or indexNot total connectivity or absolute user count
Did Cloudflare record an outage?Radar Outage Center eventsEvent methodology and coverage still matter
Did performance degrade?Network quality, latency, download, uploadMeasurement population may change
Did routing change?BGP announcements, route leaks, hijacks, AS dataRouting events do not prove user impact alone
Was one network affected?ASN-filtered viewsAggregate national data may hide local variation

Do not combine different units on one narrative without defining them. An index, percentage share, median latency, and event record answer different questions.

Seven-step workflow

  1. Record the incident hypothesis without treating it as fact.
  2. Set absolute UTC bounds, then note relevant local time.
  3. Query traffic and outage records separately.
  4. Add a matched historical baseline and neighboring or peer comparison only when appropriate.
  5. Inspect the Researcher trace and exact API parameters.
  6. Corroborate timing and cause outside Radar.
  7. Publish an evidence table that separates observation, inference, and unknowns.

If the incident is developing, use repeatable windows and record retrieval times. Later API data or event annotations may differ from the first snapshot.

Evidence table template

ClaimRadar evidenceIndependent evidenceConfidence
Traffic declinedMetric, location, timestamps, valuesSecond measurement sourceObserved / confirmed
Outage windowRadar event recordOperator or official timelineProvisional / confirmed
Likely causeNo causal proof from traffic aloneRouting, status, reports, statementsUnknown / supported
RecoveryReturn toward defined baselineUser and operator evidencePartial / sustained

Use precise language: “Radar’s HTTP traffic index fell” is stronger than “the country went offline” unless the broader claim is independently supported.

Publication and archival checks

  • Preserve the chart or API result; shared Researcher links expire after 30 days.
  • Define the Radar metric and Cloudflare vantage point.
  • State missing periods and abnormal baselines.
  • Link or cite corroborating sources near causal claims.
  • Recheck timestamps, timezones, labels, and annotations.
  • Confirm current Radar data licensing and attribution requirements.
  • Avoid exposing sensitive operational details that create new risk.

Use the Radar research planner to generate a bounded prompt and evidence checklist.

Frequently asked questions

Can Radar Researcher identify an Internet outage?

It can query Radar outage and traffic datasets and explain observed patterns. Treat the result as Cloudflare-vantage evidence and corroborate the event and cause independently.

What should an outage prompt include?

Include the location or network, absolute start and end times, timezone, metric, comparison baseline, and the specific question to answer.

Does a traffic drop prove a shutdown?

No. A drop is an observation in a dataset. Attribution may require routing data, provider status, official statements, local reports, other measurement networks, and expert review.

What evidence should I save?

Save the prompt, retrieval time, API endpoint and parameters, result or chart, trace, definitions, caveats, corroborating sources, and final reviewer decision.

Continue by task

Official sources

Source check: August 8, 2026. Verify current datasets, definitions, event methodology, API parameters, licensing, and beta behavior before publication.