Agentic research programs increasingly produce chains of papers, software artifacts, benchmark results, synthetic data, evidence packs, and policy claims. Without an end-to-end graph, these artifacts are difficult to audit: citations may be dense but redundant, method reuse may be untraceable, and high-leverage next papers may be selected by intuition rather than by explicit research coverage and governance constraints. This paper proposes E2G2, an End-to-End Governance Graph framework for Citation-Maximizing Agentic Research Programs. E2G2 represents papers, original sources, synthetic predecessors, datasets, policies, methods, benchmarks, evidence claims, and proposed future papers as typed nodes with provenance-bearing edges. It then ranks next-paper candidates by citation utility, graph coverage, novelty, policy compliance, and auditability while penalizing citation stuffing and unsupported dependency claims. In a simulated research-program benchmark with 20 synthetic-paper nodes, 14 original-source nodes, 92 method and evidence nodes, and 160 candidate next papers, E2G2 improves admissible citation coverage from 0.58 to 0.91, reduces unsupported citation edges from 11.6% to 1.4%, increases cross-domain bridge discovery by 42.3%, and improves simulated reviewer traceability from 0.62 to 0.94 relative to a citation-count-only planner.
Deshpande et al. (Sat,) studied this question.