The Absorption Problem in Software-Planning Assessment: A Single-Valued Neutrosophic Model with Deferral

Abstract

Planning the software production process is a knowledge-intensive activity whose assessment relies on expert, linguistic and frequently conflicting indicators. Deterministic and fuzzy models compress each assessment into a single membership value and therefore cannot distinguish ignorance from contradiction: when an indicator is simultaneously 'good' and 'bad', the conflict is absorbed by the projection to a scalar. We formalise this Absorption Problem and propose a Single-Valued Neutrosophic (SVN) model that represents each indicator as a truth–indeterminacy–falsity triple, aggregates them with an SVN weighted average under AHP-derived weights, computes a structural-indeterminacy operator from inter-indicator disagreement, and applies a deferral rule that routes high-indeterminacy projects to expert review. On a real project-management dataset (n = 199), 78.4% of projects show internally contradictory indicators, and the deployed fuzzy system concentrates its error there (32.7% vs 2.3%; χ²(1) = 14.57, p < 0.001). Gating on structural indeterminacy raises the reliability of the auto-decided segment from 73.9% to 97.7%. We report that machine-learning models match the accuracy; the contribution is therefore representational and epistemic — making contradiction explicit and auditable and enabling principled deferral — rather than predictive superiority.

https://doi.org/10.5281/zenodo.22292324