Reliability report
Does it matter where you monitor from?
We probed 190 widely used services from all ten of our probe regions. Availability barely moved between regions. Response time more than doubled.
Every monitoring vendor tells you to check from multiple regions. Almost none of them show you what difference it actually makes. So we measured it: 198 widely used services, probed from 10 regions, 8,906,064 measurements between 2026-08-08 and 2026-10-04.
The short answer: it barely changes what you conclude about uptime, and it completely changes what you conclude about speed. Availability across the 10 regions spanned 0.319 percentage points. Average response time varied by 2.7×.
What each region saw
| Region | Availability | Avg response | p95 response | Measurements |
|---|---|---|---|---|
London, UK lhr | 99.587% | 326 ms | 532 ms | 531,452 |
Frankfurt, Europe fra | 99.666% | 336 ms | 567 ms | 1,414,384 |
San Jose, US-West sjc | 99.702% | 362 ms | 648 ms | 1,514,876 |
Ashburn, US-East iad | 99.701% | 376 ms | 613 ms | 1,528,135 |
Chicago, US-Central ord | 99.608% | 415 ms | 713 ms | 532,094 |
Toronto, Canada yyz | 99.605% | 417 ms | 699 ms | 533,914 |
Tokyo, Asia-Pacific nrt | 99.662% | 636 ms | 1,057 ms | 1,296,648 |
São Paulo, South America gru | 99.383% | 740 ms | 1,214 ms | 526,504 |
Singapore, Southeast Asia sin | 99.528% | 817 ms | 1,292 ms | 514,650 |
Sydney, Australia syd | 99.386% | 864 ms | 1,379 ms | 513,407 |
Sorted fastest to slowest. The availability column is the share of measurements where the service answered normally.
Why this is not the finding people expect
The intuition behind multi-region monitoring is usually “a service might be down in one place and up in another.” Over this window, that is not what the data shows. The services in this catalog are mostly global platforms behind CDNs, and when they are healthy they are healthy everywhere.
What does change, sharply, is latency. If your alerting threshold is a fixed response time, the region you check from decides whether you get paged. A threshold that is comfortable from London, UK can be permanently breached from Sydney, Australia without anything being wrong.
Methodology, including what we threw out
Probes run against each service's public status endpoint from all 10 regions and are aggregated into daily buckets. Latency is weighted by sample count rather than averaged across days, because days differ in how many samples they contain and averaging the averages quietly overweights the quiet ones.
Limits worth stating. This covers 2026-08-08 to 2026-10-04, a short window, and not long enough to say anything about rare events. A single bad hour at one provider moves these numbers. It is a snapshot of a working period, not a durability ranking, and we would rather say that than imply more precision than 198 services over that window can support. The window grows as we keep measuring.
What to do with this
If you monitor from one location, the risk you are carrying is mostly not “missing an outage other regions can see.” It is a latency baseline that does not match your users, and alert thresholds tuned to the wrong number.
We check from all 10 of these regions on the free tier, which is unusual mostly because region count is the thing competitors put behind the paywall. See the live readings these aggregates are built from, or start monitoring your own service.