Multi-base station collaborative value chains have been demonstrated to have considerable potential for trajectory optimisation in human-UAV crowdsourcing systems. However, such implementations frequently exhibit compromised reliability and elevated latency, primarily stemming from heterogeneous observation states and asynchronous data collection patterns between human operators and UAVs across distributed base stations. The proposed solution in this paper is a Fully Decentralised Multi-Agent Bi-level Consensus Value-chain Offloading (FD-MABCVO) method. This method models multi-base station collaboration as a value chain optimisation problem guided by collaboration metrics. The metrics encompass transmission rate-based Age-of-Information and virtual currency-driven temporal reliability. The first-level consensus in FD-MABCVO employs a Transformer-enhanced Multi-Relational Graph Convolutional Network (TMRGCN) to extract spatio-temporal contextual features and importance weights from observations. The second-level consensus employs a blockchain-based decentralised scoring matrix and reward mechanism, integrating transaction, spatial, and environmental rewards with an adaptive balance coefficient. Experimental results demonstrate the superior efficiency of FD-MABCVO in coordinating UAVs, charging stations, and personnel across real-world scenarios at KAIST, NCSU, and Purdue.