Can you zoom in closer and also check the csv via the graphs page?
You are right that shouldn’t happen with the whaccumulator, this is the code for that process.
Knowing why or how those datapoints are missing from the other feed might also shed some light on the events at the time, can you see if there are any clues in emonhub.log and if still none the wiser, add another feed in front of these 2 on the same input and make it a phptimeseries feed, you can delete it again in a few days or when you get to the bottom of the issue. PHPtimeseries will faithfully record all data points with the timestamp, so you can see exactly what data points arrived and when to get a clearer picture than looking at the fixed interval feeds.
[EDIT}
When I count pulses I tend to use a “Total pulse count to pulse increment” followed by an “whAccumulator”, it uses more disk space but it is IMO more dependable, more accurate and far easier to debug pulse counting because it is simpler code and the first process retains the incoming data to a feed (not the increment) and the second one records the new value (last value + increment passed from first process).