<p>Exploratory item-decision audit.</p>

This study evaluated the psychometric performance of the 28-item General Health Questionnaire (GHQ-28) in an archived South African recidivist sample. Data from 385 respondents were examined using item descriptives, corrected item-total correlations, confirmatory factor analysis, internal-consistency estimates, exploratory network analysis, and post hoc reduced-item sensitivity models; the network analysis used 327 complete cases. Among the full-scale models, the correlated four-factor model showed the best relative fit, chi-square(344) = 543.81, CFI = .944, TLI = .939, RMSEA = .042, and SRMR = .105. The elevated SRMR and several negative or weak loadings, however, indicated residual misfit and possible item-direction or scoring problems. Cronbach’s alpha was.747 for the total score and ranged from.403 to.733 across the four domains. Anxiety/Insomnia and Severe Depression performed more consistently than Somatic Symptoms and Social Dysfunction. Network results were mixed rather than uniformly convergent with the latent-variable findings. Reduced-item models produced better descriptive fit indices, but they used different item sets, were data-driven, and did not establish a validated short form. Because the original item labels and scoring key were not available for independent verification, domain-level interpretations, especially for Somatic Symptoms and Social Dysfunction, remain provisional. The results provide preliminary support for a multidimensional representation of distress while underscoring the need to verify item direction, reproduce the analyses from the archived code, and replicate the findings in independent correctional samples.