Context
Peer-reviewed research by Hirsch and colleagues evaluates citation faithfulness in three open-source multi-agent deep-research systems using local diagnostics of agent invocations. In AI-Q, the authors attribute 84.7% of final-report errors to the orchestrator; roughly 31% of those orchestrator errors are hallucinations and the remainder citation mistakes.
What changed
Introduces a four-category diagnostic taxonomy and agent-level attribution method for hallucinations, uncited input reliance, uncited output, and insufficient citations. The authors report two interventions that increase citation recall by 5% in their evaluation without degrading output quality.
Why it matters
Citation visibility can fail after retrieval: evidence can be discovered and propagated between agents yet lose attribution during orchestration or synthesis. This makes citation survival distinct from initial source retrieval.
Visibility OS interpretation
Visibility OS interpretation: treat citation propagation through multi-agent workflows as a separate stage of the discovery-to-answer funnel. Do not infer that an uncited source was never retrieved.
What remains uncertain
Will the observed error distribution generalize beyond AI-Q and the three tested open-source systems? How do commercial research agents preserve citations across orchestration steps? Does improving citation recall translate to improved user attribution?