Crash-Retrieval Provenance-Independence Analysis (2026).

A leave-the-cluster-out network test on graph.json (scripts/analyze_provenance_independence.py). For each of 14 modern crash-retrieval/reverse-engineering core claims, trace source-attribution edges to any INDEPENDENT primary document; then remove the central provenance cluster (Harry Reid funding -> Robert Bigelow/BAASS -> AAWSAP/AATIP -> Elizondo/Lacatski/Puthoff/Eric Davis -> To The Stars Academy/Mellon -> Grusch) and retest. Result: 0/14 retain an independent primary-document path (100% cluster-dependent); convergence hub = legacy_program (13 sources). HONEST LIMITS: (1) this measures the graph structure, which reflects but does not replace the primary record; (2) that alleged-tier claims lack independent primary docs is partly definitional (they are testimony) - so the finding is NOT proof the claims are false; (3) the load-bearing, falsifiable result is the TOPOLOGY - a single interconnected source community, not independent witnesses - which is exactly AARO Vol 1 concluded. The claim tags (9 alleged / 4 disputed / 1 documented) already encode the weak evidentiary base; this quantifies the non-independence.