Reinforcement learning for adaptive supplier sustainability scoring and sourcing share adjustment in ready made garment supply chains

Global supply chains face increasing legal, commercial, and stakeholder pressure to verify supplier sustainability rather than rely only on supplier claims. This study develops a safety-gated tabular Q-learning framework for adaptive supplier scoring, limited sourcing-share adjustment, and audit targeting in a stylized Bangladesh ready-made garment sourcing portfolio. The simulation distinguishes direct exporters, subcontractors, and small or informal suppliers; represents safety, labour, and environmental performance separately; prohibits rewards when verified safety or labour falls below 0.60; and imposes a minimum annual audit. Three independently trained agents were evaluated in 75 paired out-of-sample replications against a contextual bandit, a risk-based adaptive heuristic, a static threshold, a combined scheduled-audit-and-threshold policy, scheduled auditing, and random action. Q-learning achieved mean true compliance of 0.862 compared with 0.820 under the static threshold, increased the share of suppliers at or above 0.80 from 58.2 to 81.3%, and reduced mean severe incidents from 3.29 to 1.71. It used 167 audits compared with 362 for the combined scheduled-audit-and-threshold policy, a 53.8% reduction, while mean procurement cost was 1.1% higher than under the static threshold. The contextual bandit produced lower compliance (0.742) and more incidents (2.96), supporting the value of delayed-action learning. The risk-based heuristic produced statistically similar mean compliance but required 55 additional audits and yielded a narrower compliant-supplier share. Paired tests, effect sizes, convergence diagnostics, supplier-stratum results, credit-assignment ablation, time-state testing, and multi-parameter robustness analyses are reported. The findings remain simulation evidence rather than factory-level validation, but they provide a reproducible and technically strengthened test of adaptive supplier governance.