Okay, so some agreements:

• We agreed to strip the discussion about spaces off the common derivatives discussion (BEP003) to then report back when we have something concrete.
• We looked into the need of splitting the concepts of "space" (as defined by the scanner's or some other way of coordinate system) and "sampling" (a discretized space: images or surfaces)
• We started to work on how to specify spaces, as you can see in the working draft. Basically, they need to describe what is the image that was used as reference. This also diverted us to talk about how you specify a template space, and considered several options.
• We started to re-work the spaces tables from BEP003
• We also saw necessary to define a "universal" transform specification that will be implemented in a software tool which purpose is 1) understanding (read/write) major formats of linear/nonlinear transforms; and 2) apply transforms to images and surfaces/point sets. The seed of this WIP is here: https://github.com/nipy/nibabel/pull/656
• We roughly agreed that the easiest way of writing this "universal" transforms on files is HDF5 (although, this would not follow ITK's current h5 format).
• We started with how these files will be named. They will have a mandatory json file that should specify what method was used to calculate the corresponding transform.
• We tried to tie together the many concepts software tools use to denote the same entities: "reference", "fixed" image and "moving", "target", "input" image. We initially agreed that a possible option is "from" and "to".