continuous raster display has "holes"

I mostly notice this when vector species habitat data has been prepped as continuous. The "edges" show a gradient where the "core" displays these "holes".

A more obvious example:

Weird, are the input tifs normal or do they also have holes?

The input tifs don't have those holes. The screen shots above are from the updated ECCC CH / ECCC SAR layers I converted to continuous. I noticed this behavior on the regional WTW projects. The same thing is happening in the Aspen Parkland, Boreal Fringe built-in project (Key Range Layers).

Rasters are projected on the fly to web Mercator, using the bilinear method for continuous data. I am wondering if something is happening there. Perhaps something to do if continuous layers have a low unique value count?

For continuous rasters "bilinear" is used for resample:

wheretowork/R/class_ContinuousLegend.R

Line 69 in 516af49

get_resample_method = function() {

I switched it to "ngb" and I get no holes:

I am wondering if there is a threshold between the number of unique values and bilinear causing holes?

An aside, I am liking having the continuous layers only display where there are values >0.