Seeing the decline of GitHub in Actions is technically correct, but Actions was always broken. We tried getting self-hosted runners to work super early before there was a proper ephemeral mode (just an officially unsupported race-condition-y --once flag). It sucked. That code can't produce a consistent status code, constantly failed to connect to its scheduler with obscure Azure error codes and had so many races with accepting and timing out jobs. Runners wouldn't get new jobs, jobs would sit there for an hour and then time out, runners would just die and need to be re-provisioned (we used ephemeral VMs in a GCP instance group). This is all because Actions is actually Azure DevOps Pipelines rebranded.

Compared to then this product is downright mature now. And also, there always were people at GitHub who delivered crappy products outside the core that most people working on FOSS got to see. Enterprise Cloud has a ton of sharp edges and things that make you ask "WHY" out loud. Notifications with SAML enabled are broken in countless ways (I have 0 out of 12 notifications right now), newly onboarded users are encouraged to click a "request copilot" button that sends emails to admins you can't disable, policy toggles that used to do one thing get split up and not defaulted properly. The last two in particular are just dark pattern hacks to get people to use Copilot. In an enterprise product.

I haven't used GHES, but I imagine it's worse.