I want to get a kind of driver reading the time value of the marker (or a given reference frame)
We can do this, but it's kind of a "stupid Blender trick." It's not practical (what's practical is to store times like this on the times of keyframes, not as GN inputs), but I'll detail it anyways.
Make an empty. Keyframe it at 0,0,0 at frame 0 and 1,0,0 at frame 1. Set interpolation and extrapolation to linear. Now, if we move that second keyframe, we can determine the time of that keyframe by comparing the current time to the X value of the empty:
If we have two frames we want to identify, those frames are on different objects (or at least, on different channels.) The keyframe at 0 cannot be moved.
These particular GN depend on the object reading the frame value to be located at the world origin (and be unparented.) We could do this in an arbitrary space by instead using two additional empties and comparing distances (we want the world space distance from 0,0,0 to 1,0,0 to establish the length of that vector, and then the world space distance from 0,0,0 to our measuring empty to compare.) Or, we could just do our math on an object that we knew was at the world origin, write to an attribute, and transfer that attribute to objects in arbitrary spaces.
However, this really is just a stupid trick, because they way to get what you want is just to set up your GN so that you directly keyframe show and hide, rather than trying to specify "when" with something other than a keyframe time:
We can see that the keyframed value node shows up in the dopesheet, as requested. I designed it this way because I believe that you're interested in the same values for all objects using these nodes; keyframing the value node inside of the GN node tree means that a single set of keyframes will be shared by all objects using that particular GN modifier. (If you would rather that these key-frames be per-object, you can make these inputs to the node tree instead-- yes, they still show up in the dope sheet.)
Of course, these keyframes that I've created on that value node will time shift just fine, along with everything else.
But you might notice that although keyframed GN nodes show up on the dopesheet, they don't show up on the timeline until we select the nodes in GN. We could just disable "only show selected" on the timeline, to make sure we can shift them. But, yes, we could also drive the visibility by location keyframes on some other object:
Here, I'm just using the Z location of an empty to drive visibility; that Z location is keyframed, and of course, it will time shift just fine along with everything else. The same warning (and solutions) I gave above about GN spaces apply.