Knowledge flow mechanisms and innovation ecosystem cultivation strategies in the digital transformation of advanced manufacturing clusters

How does digitally mediated knowledge flow shape innovation ecosystem cultivation in advanced manufacturing clusters? We address this question through a mixed-methods design organised around one primary and three secondary questions, with each method matched to a specific evidentiary task: grounded-theory coding identifies the constitutive elements of flow, two-way fixed-effects panel regression with lagged mediators traces how flow translates into innovation performance, and a calibrated system dynamics simulation compares alternative cultivation strategies. From thirty-eight semi-structured interviews we inductively derive a four-element framework-subject, object, medium, context-and a coupled dynamical system that formalizes how potential gradients, absorptive capacity, and tie strength jointly govern transmission rates. A three-tier cultivation pathway then translates flow patterns into ecosystem maturation through digital platforms, policy coordination, and industry-university-research integration. Drawing on a panel of nine clusters across the Yangtze River Delta, Pearl River Delta, and Chengdu-Chongqing region from 2018 to 2025, the empirical results are consistent with a positive direct association between knowledge flow and innovation performance, with absorptive capacity (≈ 38.6%) and digitalization level (≈ 24.1%) serving as plausible mediating channels under lagged specifications and bootstrapped confidence intervals. Under the calibrated simulation assumptions, synchronized hybrid strategies yield projected ecosystem health scores roughly 17 to 22% points higher than standalone interventions, suggesting non-additive complementarity between platform infrastructure and policy alignment rather than a simple sum of two levers. Findings extend absorptive capacity theory to digitally mediated regimes and provide sequencing-oriented guidance for cluster governance, while the boundary conditions-a small cluster panel, China-specific institutional context, and expert-elicited simulation parameters-warrant caution in cross-national generalization.