bool gpfalse = boost::geometry::within(gpo,db);

Since within() can take 2 different geometries with potentially different point types, I'm not sure to understand what can't be done in the code above using a mere grid point box. I.e. a grid point box can be passed as first argument of within() while passing a world point as 2nd argument. But maybe this particular code doesn't tell the full story of what you're trying to achieve?

The two types actually can't interact because the "Unique" template parameter in struct point makes each instantiation of different types, in the same way that the Boost.Units dimensional analysis library provides compile time protection against assignment of values that measure different units.

The code in my message does not compile because within() and related functions don't use tag_cast. I would like to be able to add additional functionality to differentiate different kinds of points In the same way as point_tag:

// From boost/geometry/core/tags.hpp

/// OGC Point identifying tag

struct point_tag : single_tag, pointlike_tag {};

I believe it should still be possible to further specialize point_tag without breaking things.

With respect to my specific case, perhaps I should instead be either inheriting from cartesian_tag or using the boost.units quantity_point extension.

The reason I originally expected point_tag to be extensible in this way is due to Barend's excellent posts on tag dispatching: http://barendgehrels.blogspot.com/2010/10/tag-dispatching-and-inheritance.html

Regarding the possible difference of scale between a grid point and a world point (your comment in the code), this is a different issue that probably appeals for something like a transformed view...

In the general case it could include any arbitrary transformation: scale, offset, rotation and transpose. This would make sense when points are essentially a proxy for addresses that can't be transformed as easily, such as the actual pixels referred to by (x,y) coordinates in an image.