Abstract
Background: Geosynthetic-reinforced soil (GRS) walls are primarily designed using strength-based limit-state checks, although serviceability is governed by wall deformation, which depends on reinforcement stiffness, facing, foundation, water and time. Gap: Within the sources examined, deformation, evidence quality, reliability and machine-learning (ML) validation are rarely treated together. Methods: This critical narrative review synthesises 72 bibliographically verified sources on working-stress design, soil–geosynthetic interaction, facing, numerical and seismic modelling, failures, water, durability, reliability, sustainability and ML. It is not a systematic review. Results: Two failure databases attribute 98% (171 walls) and 99% (320 walls) of failures to improper design or construction. Three ML studies of wall deformation reuse one set of 166 finite-element analyses with random splits; validation against measured walls was not identified. Contribution: A deformation-centred framework, a five-level evidence hierarchy (E1–E5) and a six-level validation scale (L0–L5) are proposed, together with a measurable research roadmap. Conclusions: Strength checks remain necessary but do not describe deformation. Reported ML accuracy should be interpreted according to its validation level.