Deploying something small without a platform
Most projects I work on fit on one machine. That is not a limitation, it is a budget decision: one server is cheaper and easier to reason about than a cluster, as long as you are honest about the failure modes.
What the checklist looks like
Before the first deploy I want four things in place: a systemd unit with restart policy, a log that survives reboot, a backup that has been restored at least once, and a health check someone actually looks at.
The unit is the easy part. The tempting mistake is to skip it and run the binary in a tmux session; that works until the machine reboots, and then nobody remembers the flags.
Config lives in version control
Everything except secrets goes into a repository. Secrets go into a file with 600 permissions and a note in the password manager describing what to do if it is lost.
Deploys should be boring
A deploy is: copy the artefact, run the config check, restart, verify. If the check fails, nothing restarts. If verification fails, the previous artefact comes back from disk. That is the whole process, and it does not need a control plane.
If a deploy needs a runbook longer than a screen, it is not a small system any more.