set :erb isn't a thing. You'll have to pass the layout each time. You can wrap the ERB helper like this (untested):

namespace '/cms' do
  helpers do
    def erb(template, options={}, locals={}, &block)
      options[:layout] ||= :something
      super(template, options, locals, &block)
    end
  end

  get '' do
    erb :dashboard
  end
end

Oh, but it is a thing. See this right here. You can set :<engine>, <options_hash> and it checks for the :layout option here.

Ah, it's not an official setting, but you're right, render will use it if it exists. Good catch.

I don't know why almost all settings are prohibited in namespaces. I've run into this issue myself and the workarounds are... clumsy. Perhaps someone else can chime in and explain the historical reasoning behind this decision.

Without changing the code in sinatra-namespace, and to resolve this, I can set :views to different things depending on the namespace. Some template duplication aside (does tilt play well with symlinks, you think?), it should solve my use case decently.

The question about only allowing set :views still stands though. I'm not familiar enough with Sinatra internals to know whether there are any actual issues, and sinatra-contrib has some ancient code by people that haven't been around for years, so who knows. Does this stuff have any tests?

Is the erb helper solution not an option for some reason? Tilt should be fine with symlinks, but Git (and other version control systems) not so much.

Agreed about sinatra-contrib. I and other contributors have been trying to chip away at some of the bit-rot over the past months. Code coverage is pretty good, all things considered.

Is the erb helper solution not an option for some reason?

No, it'd work just fine. It's just very clumsy. Not anyone's fault, since the code we're working around is just a little weird.

Tilt should be fine with symlinks, but Git (and other version control systems) not so much.

Agreed, I just looked into it, and git doesn't handle project-relative symlinks in any way, which wouldn't work unless everyone stores their git clones in the same path.

Agreed about sinatra-contrib. I and other contributors have been trying to chip away at some of the bit-rot over the past months. Code coverage is pretty good, all things considered.

I might see if I can just "expand" the allowed settings and see if anything breaks. It might just be fine, who knows.

I just did a minor refactor of that section, and all tests still pass. I'm putting up a PR so others can give it a test drive.

Done in #1255