A connect right after another network came up could fork a new wpa_supplicant before the killed one released wlan0 and its ctrl socket; the new daemon then failed to start and wpa_cli reported an empty state for the whole 60s poll (seen on a CULTUREHUB LA reconnect after Guest). Now waits 300ms after killall, then confirms the ctrl socket appears and restarts once if not. The spaced SSID itself was fine — quoted in wpa.conf, kept whole by the scan parser, never shell-interpolated. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>