may help to resolve this tension between granularity and contextualizability. For instance, graph-model and hypertext affordances like hyperlinking or transclusion might enable scientists to hide "extraneous details" (to facilitate compression) without destructively blocking future reusers from obtaining necessary contextual details for reuse.[^6]

Hypothesized collective benefits: Reduced overhead, and enhanced creative reuse

Discourse graphs (or parts thereof) could also significantly reduce the overhead to synthesis through reuse and repurposing over time, across projects, and potentially even across people.[^7] For example, imagine collaborators sharing discourse graphs with each other, rather than simple documents full of unstructured notes, to speed up the process of working towards shared mental models and identifying productive areas of divergence; or a lab onboarding new researchers not with long reading lists, but with discourse graph subsets they can build on over time. How much effort could be reduced if this were a reality?

The same affordances of discourse graphs around granularity and contextualizability that are hypothesized to augment individual synthesis should also facilitate exploration and reuse of an evidence collection that was created by someone else, or by oneself in the past. For example, granular representation of scientific ideas at the claim level is a much better theoretical match for the kinds of queries that scientists want to ask of an evidence collection during synthesis.¹

These claims may also be more to the level of processing required to be understood and reused by others, compared to raw annotations and marginalia.[^9] Also, ambiguity around concepts can be a significant barrier to reuse across knowledge boundaries. For example, keyword search is only really useful when there is a stable, shared understanding of ontology:[^10] this condition is almost certainly not present when [[crossing knowledge boundaries]][^11] and perhaps not even within fields of study with [[significant ongoing controversy]] amongst different schools of thought [^12]

In these settings, judging that two things are "the same" is problematic and difficult task; doing so without engagement with context can sometimes introduce more destructive ambiguity, not less, a hard-won lesson from the history of Semantic Web,[^13] ontology[^14] and classification efforts.[^15] A discourse-centric graph that embeds concepts in discourse contexts, traversing through networks of contextual details (such as authors, measures, contexts), and perhaps augmented by formal concepts as hooks, may be a better match for exploring ideas across knowledge boundaries. Further, although in many instances of knowledge reuse, contextual details tend to vary substantially across reuse tasks,[^16] there might be sufficient overlap of useful contextual details (e.g., participant information, study context) that remain stable across reuse tasks.[^17]