Reference: balanced logistic regression with 17 predictors. Baseline uses basic literacy ages 5+ minus functional literacy ages 10–64. Common-age sensitivity changes only that gap to ages 10–64; it retains basic literacy ages 5+ as a separate predictor. Historical Paper v8 uses a different unweighted specification and is not displayed here.

Populations and units: basic literacy ages 5+; functional literacy and comprehension ages 10–64; reading, writing and computation valid responses ages 10+. Internet use and applicable weekly activities cover ages 10–64. Skill scores are custom scores among internet users aged 10–64. Scores and indices /100 are not percentages of people who are vulnerable.

Clustering: the baseline uses three standardized indices (literacy, digital and their combined index): 24 high, 67 moderate and 27 low domains. Removing the redundant combined index changes 9/118 profiles (20 high, 66 moderate, 32 low); this is separate from the corrected-gap sensitivity, which does not change clusters.

Evaluation: five-fold mean metrics measure internal reconstruction of cluster-derived labels, not independent prediction of real-world vulnerability. SHAP explains the full-data fitted model in log-odds, with all 118 domains as background; it does not measure causal effects. Functional literacy and comprehension are identical predictors, and correlated contributions are not independent effects.

Limits: survey-design confidence intervals have not been estimated; small score or rank differences are not established as statistically significant. Study boundaries contain small overlap flags under review and are not certified legal boundaries. Display edits do not recompute the research results.