I think the downtime calculation is described on these 2 posts below.
In a nutshell, what I understand is:
- the downtime score is based on audits (i.e. audits requests sent by the satellites to the node)
- the node may receive a varying number of audits from satellites
- then, for a given node, the score may be different depending on the satellite we are considering.
Please correct me or complete if I’m wrong or missing something. 
(would be really nice to have a topic on this on https://support.storj.io/)