From the link you provided I read:
"If the coordinates are not covered by the SRTM data we will temporarily fallback to using OpenStreetMap's OpenElevation API. We have found the data in this API to be extremely unreliable, therefore we will only use it as temporary data. Immediately after adding the data from the OpenElevation API we will add a new elevation job into our queues. This time it's flagged to use the Google elevation API.
The Google elevation API has API limits that we have to respect, therefore there might be another few minutes of delay before we actually fetch this data. We are trying to bulk multiple locations together to help us with the rate limits. Meanwhile we are relying on the temporary OpenElevation data added before."
This sounds like the reason for the problem you mention: "On first day after publication all was ok, but on next day it showed average level of a county - difference is about 300 metres in height."
Besides this, I think the explanation in the link is quite clear. I think your question is more about how exactly the SRTM data is acquired.