In the operational, strategic and tactical decision-making problems of the agri-food supply chain, the perishable nature of the commodities can represent a particular complexity problem. It is, therefore, appropriate to consider decision support tools that take into account the characteristics of the products, the needs and the requirements of producers, sellers and consumers. This paper presents a green vehicle routing model for fresh agricultural product distribution and designs an adaptive hybrid nutcracker optimization algorithm (AH-NOA) based on k-means clustering to solve the problem. In the process, the AH-NOA uses the CW algorithm to increase population diversity and adds genetic operators and local search operators to enhance the global search ability for nutcracker optimization. Finally, the experimental data show that the proposed approaches effectively avoid local optima, promote population diversity and reduce total costs and carbon emission costs.

The feed conversion ratio (FCR) is a key indicator of efficiency and sustainability in aquaculture, yet existing evidence on its farm-level determinants remains dominated by nutritional experiments, with little rigorous identification of the causal effects of managerial inputs or their heterogeneity across production regimes. Using farm-level survey data from 222 Litopenaeus vannamei farms in four major Chinese producing provinces over 2016–2023 (n = 382 farm-cycle observations), we applied propensity score matching (PSM) with subsample heterogeneity analysis, validated by Rosenbaum bounds and OLS interaction tests, to identify the effects of branded-feed adoption and technical-service access on FCR. Four findings emerge. First, in the pooled sample, both branded feed (ATT = −0.31, p < 0.05) and technical services (ATT = −0.24, p < 0.05) significantly reduce FCR. Second, among traditional pond farmers, branded feed yields a substantial efficiency gain (ATT = −0.528, p < 0.01), whereas technical services do not. Third, among industrialized farmers, the pattern reverses: technical services significantly reduce FCR (ATT = −0.376, p < 0.01), whereas branded feed has no detectable effect. Fourth, Rosenbaum bounds confirm that the significant ATTs withstand hidden bias up to Γ ≈ 2.8–3.0, and OLS interaction tests independently replicate the reversal. The evidence supports a complementarity-matching hypothesis: the marginal value of a modern input depends on whether it relieves the binding constraint of the production regime. The findings inform a differentiated extension policy for sustainable shrimp aquaculture.