The integration works. The question is whether it still will in a year.

Ask who controls the interface

A published, documented, versioned API that the provider's own paying customers depend on is a stable thing to build against. Their commercial interest is aligned with not breaking it.

A scraped page, an undocumented endpoint, or a file export whose format nobody has committed to is not an integration. It is an arrangement that works until the day it does not, and it will fail without warning on somebody else's release schedule.

Five questions before committing

1. Is the interface documented and versioned, with a stated deprecation policy?
2. What are the rate limits, and what happens to your process when you hit them?
3. How does authentication renew, and what breaks when a token or credential expires?
4. What happens when the provider is down? Does your process queue, or fail?
5. Who notices when it stops working? An integration nobody monitors fails silently, and silent failures are found by customers.

Design for the second-worst case

The worst case, a provider shutting down permanently, is rare and obvious. The second-worst is far more common and much harder: intermittent failure, partial data, or an interface that returns success while silently dropping records.

Any integration handling money or stock should reconcile rather than trust. Count what you sent, count what arrived, and alert on a difference. Systems that assume success are the ones that quietly diverge for six weeks.

Why we mark most of ours as on request

Applying the questions above to our own integration list is the reason nearly all of them carry a conservative status rather than being described as available.

A logo on a marketing page is not evidence that a connection works in production for your configuration. We would rather scope it with you and be right than list it and be optimistic.