Time-Invariant Models, Time-Varying Effects: The Reversal of Network Embeddedness across Professional Careers

Research on the career network treats the effect of the relational structure on outcomes as invariant with time.Among 495 elite Korean artists over seven decades (1929--2002), the association between network stability and career plateau risk is absent early but strengthens monotonically from year~10 to year~20, while network size remains protective throughout.The pooled coefficient averages over these stages, describing a statistical artifact that corresponds to no actual career phase, regardless of causal direction.Bidirectional feedback between network consolidation and productivity decline intensifies over career time, constraining all admissible causal models.Quasi-experimental evidence from institution closures supports \textit{evaluative redundancy---diminishing} returns from repeated endorsements by the same gatekeepers---as a forward-acting pathway.The association escalates across the achievement distribution.A pilot replication in the film industry supports the phase reversal: a data-driven change-point model identifies overlapping thresholds in both datasets, and a structural break test confirms the pattern.Replication code and data: https://github.com/joonhyungbae/TITV