LatencyRadarBot

1. What it is

Latency Radar measures how long websites and APIs take to respond from five locations around the world. LatencyRadarBot is the software that makes those requests.

Each request is an ordinary HTTP request for one URL. We time how long it takes to find the server, connect, set up TLS, receive the first byte, and receive the response. We record the timings, the status code, the connected IP, and a small set of reviewed CDN/edge headers so we can tell which network served the response.

The response body is read only so the timing is complete. It is not stored, indexed, or republished. We do not keep cookies, credentials, arbitrary headers, or page content.

2. What it is not

• Not a crawler or scraper. We request only the specific URL that was configured or submitted. We never follow links, discover pages, build an index, or extract content. Nothing from your pages is retained or reused.
• Not an LLM capability benchmark. For AI inference APIs we measure one thing: the time until the first visible token arrives. We send a tiny fixed prompt with a very small output limit through a paid API key. We do not evaluate, score, store, compare, or publish model outputs, and we do not use any output to train, fine-tune, or build models. Timing is the only measurement, and the methodology is published so anyone can replicate it.
• Not a load test. One request at a time per location, with a pause between requests. A full day of scheduled measurement against one target is a few dozen requests in total.

3. Two kinds of traffic

User-requested checks

Someone using latencyradar.com or its API asked us to test an endpoint. A quick check is one request per location; a deeper check is 20 requests per location, one at a time. These are one-off runs started by a person, who agrees in our terms that they are authorised to test that endpoint.

Scheduled reference collection

We also measure a curated catalogue of well-known public websites and APIs on a schedule, to publish the benchmark reports under /latency/. The cadence for each target is:

• Websites and APIs: five requests per location once a day, and twenty requests per location once a week.
• AI inference routes: three to five requests per location on Monday, Wednesday and Friday, and twenty per location every two weeks.
• Start times are randomised within a window, so the traffic does not arrive at the same second each day.

4. Where the requests come from

All measurement requests originate from these hosts. The region value in the User-Agent tells you which one. This list is updated whenever a location is added or changed.

5. What happens if you block us

A 401, 403 or 429 response is treated as a signal, not an error to retry through. It is recorded as a finding (that a service limits third-party measurement), the target is paused automatically after repeated blocks, and we do not retry into rate limits.

We do not crawl, so robots.txt crawl rules do not map cleanly onto what we do and we do not currently read it. Returning 403 to the User-Agent above, or emailing us, both work.

6. How to opt out

Email support@latencyradar.com with the hostname. We will remove it from the reference catalogue so it is no longer measured on a schedule, and on request we will refuse further checks against it.

If you operate a service we publish a benchmark report about and want to check the methodology or replicate a measurement, the same address reaches the people who run it.