Note that connection errors that happen during finetuning make finetuning information significantly distorted, so it is by-design behaviour to restart finetuning process if connection error has been detected.
Unfortunately, it is not possible to detect at which stage connection became invalid, if this produced extra CPU load or introduced scheduling pauses.
You can perform finetuning at the pool with high connection stability and use AV parameter with any other pool, because of mining algo is pool neutral.
yvg1900