It is very interesting to hear that you are exploring this. This is actually missing in almost all of the tools within TFT space. I think assigning weighting to references is a very specific case of attaching data/properties to links/edges. So I would argue that reimagining references based on a more flexible/generic property graph approach would provide the most benefit. Even if the details of implementing properties on a references is a big ask, I think building on a model that allows for that in the future is a better approach.
I think having the ability to transclude/embed blocks of content across entities/documents is one of the major shortcomings of fibery at the moment, if you are using it beyond a database/airtable substitute (i.e. if you want to use fibery to its full potential). I find myself copy and pasting content repeatedly just to ensure the new item (e.g. meeting notes, specifications, ideas list, …) has some of the prior context in it and then having to spend time keeping the multiple versions in sync or keeping track of the latest version. Transclusion neatly resolves this.
However, I do think any effort towards transclusion should ensure that it doesn’t require the source content to be formatted in a particular way (a priori) or require modifications to it when transcluding. Both of these would add a lot of friction to the process. I discussed the issue here (responding to how early versions of blocks looked in fibery). I do think Obsidian or Roam’s approach to each bullet, line being a block/node that can be transcluded is the way to go long term.
I think this issue has come up a couple of times. I do find it immensely helpful in some context to be able to edit original content in another context. However, I do understand the concerns that it might be dangerous to do in other context. I think an easy way of dealing with this might be through settings/customization. It could be that transclusion by default just reproduces (or embeds) the content and the user/admin would have to intentionally allow editing of transcluded content (out-of-context).