This counts how many Packet-of-Disconnect failure notifications the gateway has sent, each one reporting that a requested disconnect could not be carried out. A nonzero and climbing count here means the gateway is failing to comply with the RADIUS server's disconnect instructions, often because it can't match the accounting session ID in the POD request to a live session, which leaves sessions the billing system already considers closed still active and consuming resources on the gateway. A spike right after a mass subscriber re-provisioning would suggest stale RADIUS session IDs are being sent that no longer match any live session on the gateway.