Make sense of it

Cross-Source Synthesis

One answer drawn from everything you know.

Cross-Source Synthesis takes a natural-language question and produces a grounded prose answer by integrating evidence across multiple documents or knowledge sources — not just the closest match. Use it when the answer lives across several sources and you need a coherent, cited response rather than a list of passages. It's the primitive behind knowledge-base Q&A and multi-doc research tools.

Shape

Operational dimensions

Human co-pilot

Person and system work side-by-side.

On demand

Fires when a user asks.

Medium data gravity

Holds working state that compounds over runs.

Read-only inbound

Consumes external data; does not write back.

Inputs

natural-language query
retrieved passage set spanning multiple corpus sources
grounding policy (citation style, scope constraints)
optional context (user role, prior conversation turns)

Outputs

synthesised prose answer integrating evidence from multiple sources
source citations with passage-level provenance
confidence or coverage signal indicating source breadth
optional: gaps flagged where corpus coverage is thin

Mechanism

Answers a natural-language query by retrieving relevant passages across multiple corpus sources and synthesising a grounded prose answer that integrates evidence from more than one source.

Why this is a primitive

Cannot be decomposed: the synthesis step (read N retrieved passages, reconcile, compose a single grounded answer with citations) is one operation that goes beyond retrieval. Pure retrieval returns the passages and stops — that is `find-your-way-around`. Synthesis is the irreducible derive-an-answer-across-sources move. The retrieval step IS used (composed_of at the buyer layer) but the make-sense primitive is the synthesis itself.

Where it shows up

Professional services firm — consultant asks 'what does our methodology say about change management in public sector?' and gets a grounded answer drawn across five internal playbooks

EdTech platform — learner asks a concept question and receives an explanation synthesised from curriculum guides, glossary, and worked examples

Healthcare operator — clinical team asks about contraindication across multiple drug monographs and receives a reconciled summary with citations

Legal team — associate queries 'what are our standard indemnity positions across service agreements?' and gets an integrated answer from the contracts corpus

Related primitives

Composes with

Corpus Retrieval →Faceted Filtering & Browse →In-Source Navigation →Descriptive Aggregation →Long-Form Prose Authoring →