Pimp My IDE / garage dispatch
Back to the garage
October 6, 2026 | testing / public infrastructure / network proof

The example is not your test server.

A reserved name is safe to print in documentation. Its courtesy web page is not a production dependency. Test the layer you mean against an endpoint you control.

Name the question before you pick the host.

Reserved does not mean reliable.

RFC 2606 reserves example.com, example.net, and example.org for examples. It also reserves .test for testing DNS-related code, .invalid for names that should fail, and .localhost for loopback use.[2]

That reservation prevents naming conflicts. It does not promise an HTTP response. IANA says the web service on the example hosts is best effort and should not support production applications.[1]

A safe name in a README is not a service-level agreement.

The redesign exposed the hidden dependency.

DebugBear traced the recent example.com change. IANA split the page into basic HTML and a separate JavaScript file because automated traffic often skips the script. That cuts bandwidth on a heavily requested courtesy endpoint. The new page also says not to rely on it for testing or monitoring.[3]

The Hacker News thread quickly filled with guesses about broken tests and connectivity checks. The thread is useful as a warning, not proof that a specific test failed. It shows how easily a familiar public page becomes an unstated fixture.[4]

Split the network question into layers.

"Is the internet up?" is too broad for one green light. A DNS lookup can work while TLS fails. A TCP connection can work while the application returns the wrong body. A content assertion can fail after a harmless redesign while the service remains healthy.

Write the exact question first. Use an endpoint your team owns for service checks. Force the failure you expect the check to catch. Keep documentation examples on reserved names, but do not confuse the example with the fixture.

A test needs a failure you can schedule.

The breaker below drafts a test contract. It does not send traffic. The selected layer changes the target requirement and the observable. Four interlocks cover ownership, a controlled failure, an exact assertion, and cleanup.

A complete control bank means the template has all four sections. It does not mean the endpoint exists or the test passed.

Interactive makeover / network test contract

Courtesy endpoint breaker

Traditional purpose replaced: one generic ping or page fetch. Better version: choose the layer, close four independent interlocks, and copy a contract that keeps the target requirement and failure drill visible.

QuestionTargetFailureReceipt
What are you testing?
Close the plan interlocks

These controls select sections for a plan. They do not inspect a host, create an endpoint, or run a failure drill.

Two narrow assertionsDNS: A record equals the fixture IP.HTTP: status is 204 and Cache-Control is no-store.
0 of 4 interlocks selectedPlan only

Choose a controlled DNS zone, then define the expected record.

DNS layer selected. No interlocks are selected.

Sources read

Source log and evidence boundary
  1. IANA, "Example Domains", revised recently and read October 6, 2026. IANA says the domains are maintained for documentation and that their web service is best effort, not a production dependency.
  2. RFC 2606, "Reserved Top Level DNS Names", published June 1999 and read October 6, 2026. It defines the reserved testing, example, invalid, and localhost names and explains the conflict they prevent.
  3. DebugBear, "Example.com Just Launched The Biggest Redesign In Decades", published September 30, updated October 2, and read October 6, 2026. It documents the page history, the recent split between basic HTML and added JavaScript, and IANA's stated bandwidth reason.
  4. Hacker News item 49971921, read October 6, 2026. The discussion raised possible test breakage and connectivity-check habits. It is discovery context and user opinion, not evidence that a named production system failed.

Evidence boundary. The component generates text in this browser. It sends no request and validates no target. The suggested questions are editorial guidance. A selected interlock records intent only. Run the real test against infrastructure you control and save its output before making a reliability claim.