synthesis is fundamentally about creating a new whole out components (strike1983types). This means that synthesizers need to be able to compose individual ideas (evidence, hypotheses, concepts, claims) into a larger conceptually integrated understanding, such as a theory or argument.
In sensemaking, this is known as constructing a new understanding or Schema from raw data in order to aid some downstream task, like decision-making (russellCostStructureSensemaking1993, pirolliSensemakingProcessLeverage2005). For example, intelligence analysts might need to recommend a course of action for a state's military; to do this, they might need to extract relevant evidence from different pieces of data, weigh how they provide warrants for different hypothesized outcomes, and then connect these outcomes to possible courses of action (or sequences thereof) to enable desired outcomes and/or prevent bad outcomes.
(faisalClassificationSensemakingRepresentations2009) identifies a finite set of six "types" of representations / schemas that people construct during sensemaking: 1) spatial, 2) argumentational, 3) faceted, 4) hierarchical, 5) sequential, and 6) networked.
We can also look at examples of intellectual acts that do not seem to produce something more than the sum of their parts.
Classic undifferentiated list, as noted in @holbrookLevelsSuccessUse2008 and others.
So, idea of "parts, not whole"? Assemblage? Can't transcend the sum of the parts. Or might not even have the sum.
The idea of a conceptually new whole construction is in contrast with what we often see in related work sections and literature reviews, which are much more about assemblage, often listing in chronological or author-order, what things have been done, what topics have been covered, with little to no synthesis into a coherent new whole.
It's not just that they're more easily done in well-structured domains where we have a bit more agreement.
It's also that these systematic reviews and meta-analyses are often laser-focused on just one edge in a causal model!
Granted, good reviews (especially meta-analysis) might try to estimate some of the variability, but they're rarely focused on explicating a causal model or larger whole.