CISCO ACI - overhead on multi POD and multi SITE

As far as I know, ACI-OS, NX-OS and IOS don’t take in account ethernet header (14 – 18 bytes) whereas IOS-XR includes the ethernet header when you have to estimate the right size of the MTU. Everywhere, looking about the MTU overhead on Multi POD and Multi Site implementation, I read the message: "Keep in mind when configure Spine and IPN/ISN nodes to consider an extra bytes of 50".

My question are two:
1) First, why 50 and not 54 being the multi pod and multi site implemented on a sub routed-interface on Spine (Alan tag 4) that involves so +4 bytes on top of the rest?
2) Second, why I should consider however an extra of 50 bytes instead of ONLY 36 bytes on spines (then on IPN/ISN it will depend on platform chosen and OS if add or not other 18 bytes for the L3 header overhead) if ACI-OX is not counting the L2 overhead?

Thanks for your answers
Mario