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
| Question | Candidate Radar evidence | Caveat |
|---|---|---|
| Did HTTP activity fall? | HTTP traffic time series or index | Not total connectivity or absolute user count |
| Did Cloudflare record an outage? | Radar Outage Center events | Event methodology and coverage still matter |
| Did performance degrade? | Network quality, latency, download, upload | Measurement population may change |
| Did routing change? | BGP announcements, route leaks, hijacks, AS data | Routing events do not prove user impact alone |
| Was one network affected? | ASN-filtered views | Aggregate 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
- Record the incident hypothesis without treating it as fact.
- Set absolute UTC bounds, then note relevant local time.
- Query traffic and outage records separately.
- Add a matched historical baseline and neighboring or peer comparison only when appropriate.
- Inspect the Researcher trace and exact API parameters.
- Corroborate timing and cause outside Radar.
- 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
| Claim | Radar evidence | Independent evidence | Confidence |
|---|---|---|---|
| Traffic declined | Metric, location, timestamps, values | Second measurement source | Observed / confirmed |
| Outage window | Radar event record | Operator or official timeline | Provisional / confirmed |
| Likely cause | No causal proof from traffic alone | Routing, status, reports, statements | Unknown / supported |
| Recovery | Return toward defined baseline | User and operator evidence | Partial / 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
- Radar Researcher beta access and limits
- Code Mode and MCP architecture
- Radar WebMCP browser agents
- Radar research planner
Official sources
Source check: August 8, 2026. Verify current datasets, definitions, event methodology, API parameters, licensing, and beta behavior before publication.