Well, if you have a look at help isa,

then it turns out that for isa(x,"float"),
float is not really a class,
it is interpreted as 'floating point value'
comprising classes double and single.

So further classes may represent floats.

For classes double and single, the function isfloat(x)
returns true indicating that these are floats.
If I write my own class, then I would also define isfloat(x)
returning true.

Then isa(x,"float") can determine whether it is a float
just asking isfloat(x).

I write an arithmetic with oo-means,
so overwriting isfloat, isinteger and isnumeric
for my own arithmetic classes.

In addition, i have to overwrite isa
for second argument 'float', 'integer' and 'numeric'.

I suggest to modify builtin function isa
so that for these second arguments
methods isfloat, isinteger and isnumeric are invoked.

This improves consistency and spares me overwriting function isa.