Subsystem not namespaced when substituted coefficient is 1

[hersle]

I am puzzled by this behavior:

using ModelingToolkit
using ModelingToolkit: t_nounits as t

function buggy(ϵval)
    @variables x(t)
    ϵ, = GlobalScope.(@parameters ϵ) # never namespace
    @named A = ODESystem(Equation[], t)
    @named B = ODESystem([substitute(ϵ*x ~ 0, ϵ => ϵval)], t)
    C = compose(A, B)
    return equations(C)
end

buggy(-1) # [-B₊x(t) ~ 0], as expected
buggy(0)  # [0 ~ 0], as expected
buggy(1)  # [x(t) ~ 0], but expected [B₊x(t) ~ 0]
buggy(2)  # [2B₊x(t) ~ 0], as expected

Why is the namespacing on x dropped only when ϵ => 1 is substituted?

This blocks me from programmatically generating a more complex hierarchical system.

[ChrisRackauckas]

To be perfectly honest, I'm also pretty baffled. It might have to do with the 1*x => x rule getting in the way, but I don't see why that would effect namespacing.