Questions about field collisions when GBIF builds occurrence records

[Mesibov]

In many datasets I've seen, event.txt and occurrence.txt share fields, for example eventDate. When GBIF builds occurrence records from event.txt+occurrence.txt, what happens when the "same" field is in both files? Is one field discarded? Which one?

Does GBIF check for inconsistencies in field collisions?, e.g.
eventDate = 1963-02-06 in event.txt for eventID "something2334"
eventDate = 1973-02-06 in occurrence.txt for N records with eventID "something2334"

[Mesibov]

For anyone following this issue, @timrobertson100 of GBIF has answered a related question here: tdwg/dwc-qa#201 as follows:

"In GBIF processing today, the data is pivoted to occurrences such that the fields on the event will only be used if they are null on the occurrence records. In this instance, those event properties would be dropped. There is exploratory work to bring in an event index where both fields would remain, but that is some way out."