How is configuration drift / policy enforcement handled if there's no polling interval? Like suppose a user changes a setting, does it revert back to the enforced setting, if so, when and how?
Delivery is push, not pull. Each agent holds a persistent gRPC stream to the server (mTLS). Policy changes are broadcast over that stream the moment they're released — agents apply them in real time. If the connection drops, the agent reconnects with backoff and sends its last-known revision; the server replays exactly what it missed (or a full snapshot). So there's no "check every N minutes" anywhere in the design.
Most user changes never stick in the first place. Where the desktop stack has a native lockdown mechanism, Bor uses it: dconf keys are written together with a dconf locks file, so a locked setting simply can't be changed from GNOME's UI; KDE settings are written with the Kiosk [$i] immutability marker, which KDE itself enforces; and everything else (Firefox/Chrome/Edge policies.json, polkit rules, firewalld zones) lives in root-owned files under /etc that an unprivileged user can't touch. Those applications treat managed policy as non-overridable by design.
Root-level drift is caught by inotify, not a timer. For every file it manages, the agent watches the parent directory via inotify (parent dir, so atomically-renamed replacements are caught too). If anything external modifies or deletes a managed file — a curious admin with sudo, a config-management tool, a package postinstall script — the watcher fires immediately and the agent rewrites the file from its cached policy state and re-reports compliance to the server. In practice the revert happens within milliseconds of the write, and the tamper event is visible server-side, so drift isn't just corrected — it's auditable. (The agent suppresses events from its own writes, so it doesn't fight itself.)