Prioritising AI-Augmented Requirements Engineering Phases for Legacy System Modernisation Using Symmetry-Informed SF-AHP Weighting and Spherical Fuzzy Scoring

Sequencing artificial intelligence (AI) investment across requirements engineering (RE) phases remains an open problem in legacy system modernisation, particularly when expert judgements include hesitancy. This study prioritises five phases of the AI-Augmented Requirements Engineering model using spherical fuzzy analytic hierarchy process (SF-AHP) criteria weighting followed by direct spherical fuzzy phase scoring. Experts selected linguistic terms mapped to predefined spherical fuzzy triplets; the coordinates were represented explicitly but not elicited separately. Continuous spherical aggregates were converted through a calibrated reciprocal exponential map, while linguistic labels were retained only for interpretation. The selected score factorises into an antisymmetric polarity term and a hesitancy-dependent modulation term. We show that score antisymmetry, combined with reciprocal rescaling, preserves the AHP reciprocity imposed by the reflection convention. In a banking transformation involving three experts, Defect Triage and Test Scenario Generation formed the leading tier, while TO-BE Design ranked last. The within-pair margin was 0.0079. The ordering reversed in 18 of 120 single-step perturbations and was retained in 57.5% and 55.0% of the Monte Carlo instances at the two noise levels, respectively. Ablation and sensitivity analyses identify which modelling choices affect ranking stability. The results support tier-level prioritisation rather than a strict adoption sequence.