Like many organisations, we have an editing database and a publication database, and we use Esri 1way replication to keep the versioned feature classes in sync. To complicate matters, we have geometric networks and many relationship classes as part of the replica.
We have long observed that some of the relationships do not make it across - that is feature x is related to feature y on our editing database, but they are not related on the publication database, aka the replica target. We have never been able to reproduce this, and when investigated in the past, the discrepancy counts were low enough that we didn't think it warranted too much digging.
However, we found today that at least for one subtype of one feature class, the results are pretty disastrous. There are approximately 60,000 records of this particular subtype, all of which should have a related feature. On the editing database, about 600 do not have the required related feature. These are likely data errors, but only affecting 1% of the data...we can manage that. But on the publication database, roughly 30,000 of these features are lacking their related feature! Half! And I of course expect we will find some similar ratios as we expand our analysis.
So my question for all you out there in GeoNet world - has anyone seen this before?? Any ideas as to what might be happening?