Test synthesis
Test synthesis output by asking whether it makes a claim that could not be found in any individual source - if it could be produced by listing inputs, you have aggregated rather than synthesized.
When This Fires
- After completing a synthesis session (Lay out multiple notes in parallel visual access…, Look for structural parallels across inputs during…) when evaluating your output
- When your "synthesis" feels like a summary with a fancy label
- When reviewing written work to determine whether it contributes new understanding
- Complements Look for structural parallels across inputs during… (structural parallels) with the validation test for synthesis quality
Common Failure Mode
The "both A and B say X" non-synthesis: identifying shared claims across sources without producing an emergent insight. "Both Kahneman and Ariely describe cognitive biases" is aggregation. "Kahneman's dual-process model explains why Ariely's specific biases cluster into predictable System 1 failures" is synthesis — it creates a framework that neither author stated explicitly.
The Protocol
(1) After producing synthesis output, extract your core claim — the central statement your synthesis makes. (2) Apply the single-source test: "Could I find this exact claim in any one of my input sources?" (3) If yes → you've extracted or summarized, not synthesized. Return to the inputs and look for emergent patterns (Look for structural parallels across inputs during…) rather than shared content. (4) If no → the claim is emergent. Verify it further: "Does this claim require the specific combination of these inputs to produce?" If it could have been generated from any random pair of sources, it may be too generic to be valuable synthesis. (5) The highest-value synthesis produces claims that are both emergent (not in any single source) and specific (require this particular combination of inputs, not just any inputs on the same topic).