Source-grounded synthesis

Three slides into a vendor-selection brief, the VP stopped the meeting. "This says switching costs are low. How do you know that?" The analyst had the number somewhere in his notes, but not the source. He remembered reading it, wasn't sure where, and the meeting moved on without a real answer. The recommendation technically still stood, but nobody in the room trusted it as much as they had ninety seconds earlier.

Source-grounded synthesis is what would have saved that meeting. It means combining many sources into a view where every claim keeps its evidence attached, so a conclusion that sounds plausible can also be checked, on the spot, by whoever asks.

Fluency is not evidence

Unguided AI summarization is the extreme version of the same failure. It produces fluent output that reads as authoritative while quietly severing the link to evidence. Verification practice in other fields has the same starting rule: a claim is not usable until it can be traced back to where it came from, and tracing is a separate step from finding the claim plausible (Verification Handbook).

The unit of synthesis is the grounded claim

A grounded claim has three parts: the assertion (what you are saying is true), the support (the specific passage, figure, or statement that backs it), and the qualification (the conditions, assumptions, or uncertainty attached to it).

When all three travel together, synthesis becomes inspectable. A reviewer can follow any assertion to its support and judge whether the qualification is honest. Drop the support and qualification, as most quick summaries do, and the assertion floats free. Trust then has to be granted rather than earned.

Work the disagreements, don't average them

Multiple sources rarely agree cleanly, and the instinct to produce one tidy takeaway often means averaging away the most informative part of the picture. Writing guidance for combining sources makes the same point directly: don't force a relationship between sources if there isn't one, and take note of every perspective, including the ones that disagree with your broader conclusion (Purdue OWL, Synthesizing Sources).

In practice that means:

1. Cluster claims by question, not by source. Bring everything the material says about one sub-question together.
2. Surface conflicts explicitly. When two credible sources disagree, that disagreement is a finding, recorded with both sources attached.
3. Weight by grounding, not confidence. A hedged claim backed by primary data beats a confident claim with no support.
4. Separate observation from inference. Keep what the sources say distinct from what you conclude, so a reader can challenge each independently.

Keep the trail intact through every revision

Synthesis is not a one-time act. As new sources arrive, the view should update, and the evidence links have to survive each revision. The common failure is that provenance erodes over edits: a claim that started grounded gets reworded, generalized, and eventually orphaned from its source. The fix is to treat the evidence link as part of the claim, not a footnote added at the end.

Weavu builds this in rather than leaving it to discipline alone: a capsule, its term for a synthesized view you steer over time, keeps each claim's source link attached as the capsule gets revised, and the underlying graph, a map of how claims and sources connect, shows at a glance where the support is thick and where it is still thin.

A synthesis you can defend

The test of source-grounded synthesis is simple: hand the conclusion to a skeptical colleague and ask them to check it. If they can trace each claim to its support and judge its qualifications without redoing your research, the synthesis is doing its job. That VP's question should have a thirty-second answer, not a shrug.

If you're ready to turn a set of grounded claims into a decision document, the guide on building a research brief from messy sources covers that next.