Try swapping the order of CoordinateSystemSetter and VertexCreator. I find FME accepts the CRS set operation better if there is a geometry on the feature.

I have done it, but it does not solve the issue.

Another thing to look at then is the reprojection process itself.

Transformations involving MGA2020 can be tricky. Try a multi-stage reprojection such as: LL-WGS84 > LL-GDA2020 > MGA2020-Z56 or even LL-WGS84 > LL-GDA94 > LL-GDA2020 > MGA2020-Z56.

Also the use of the CSMapReprojector can give you a little more control.

Thanks so much, it worked! 
I really appreciate the help. I’ll share my workflow here in case it’s useful for anyone else who runs into the same issue.

Firstly define the two fields as X and Y coordinate fields and then also their source coordinate system in the reader, so they come out as spatial features.  You can also do this for Excel files.

After this you only need the Reprojector or CSMapReprojector to get to your destination coordinate system, which it seems you’ve done.  As well as whatever you are testing.

​@chriswilson while this is a cleaner method of creating spatial features direct from the CSV and I would use it every time, the workspace intiially described acheived the same thing using transformers. The issue lay in the subsequent reprojection task which for tranformations involving MGA2020 can require moving through an intermediary CRS to minimise positional errors creeping in.The transformation between WGS-84 and MGA2020 Zone 56 is not well defined, if at all.