Paper 4 · Question 10

A: Inline hyperlinks in prose are fragile — the synthesis agent will rewrite, merge, and compress prose during summarisation, stripping or disconnecting links from their associated claims. Attribution must be in structured data, not embedded in prose that gets rewritten.

B: Structured claim-source mappings survive synthesis because they are data structures, not prose that gets rewritten. The synthesis agent can merge mappings from multiple subagents while preserving the link between each claim and its source. This also enables content-appropriate rendering: financial data as tables, news as prose, technical findings as lists — each with attribution intact.

C: A bibliography lists sources but does not map specific claims to specific sources. Stakeholders need to know which claim came from which document, not just which documents were consulted overall. This is document-level attribution, not claim-level provenance.

D: This adds infrastructure complexity without solving the synthesis problem. The synthesis agent still needs to merge and present findings coherently. Database references do not prevent attribution loss during the summarisation and rewriting that synthesis inherently requires.