How timeseries sync deals with double values for same timestamp?

What does it happen, if in the backing dataset there are two rows for the same series ID and same timestamp but with two different values?
Is there a predictable behavior?
Does the nature of the backing dataset (streaming or batch) influence it?

I also noticed that there is the ability to optionally configure an ingestion timestamp, is it impacting the behavior regarding double values for same timestamp?

Asking AIP Assist on the stack seem to suggest that it is random. However, it is not clear from where in the documentation it is extracting this information. I also asked about the ingestion time and it says that it does not influence the deduplication behavior and is just metadata. However it does not seem to be used or available anywhere on the read side. So I’m wondering if there is not some un documented part of it.

From the product team:

Behavior is not deterministic: A timestamp will be arbitrarily chosen since timeseries can only have 1 point per timestamp.

Even a simple latest value is not reliable as replays/re-snapshot/re-hydrations will read the 2 points for the same timestamp in potentially different orders.

And so user should not rely on any observed behavior as this might not be consistent.