Who the Algorithm Doesn't See: A Practitioner Audit for Pre-Evaluative Exclusion in High Informality Economies (Latin America)

Contemporary Artificial Intelligence Governance Frameworks have focused predominantly on downstream concerns such as bias detection, explainability, robustness, transparency, and post-deployment risk management. Yet a prior institutional question remains insufficiently examined: which populations can actually become meaningfully evaluable within an algorithmic system before any score, classification, eligibility determination, prioritization, or automated recommendation is produced.

This paper introduces the concepts of «Admissible Evidence Boundary» and «Pre-Evaluative Exclusion» to examine how algorithmic systems construct evaluability through the records, variables, signals, and documentary forms they recognize as admissible evidence. Focusing on Latin America and high informality economies, the paper analyzes how substantial portions of economically active populations may remain only partially visible within automated evaluative infrastructures despite participating continuously in markets, labor systems, and social life.

To operationalize this governance problem, the paper develops the «Velarde Pre-Evaluative Audit (VPA)», an institutional governance instrument designed to identify divergences between the population organizations intend to evaluate and the population their systems can effectively evaluate under real deployment conditions. The audit is structured across four institutional decision environments: vendor disclosure, board oversight, investment due diligence, and public procurement.

The analysis is illustrated through five Latin American cases involving social targeting, credit scoring, predictive administration, and automated classification systems: SISBEN IV in Colombia, RappiCard in Mexico, PTIS in Argentina, Registro Social de Hogares in Chile, and SISFOH in Peru.

Rather than proposing a new regulatory regime, the paper positions evaluability itself as an object of institutional oversight. It argues that governance failures may emerge not only from biased or inaccurate outputs, but also from upstream documentary conditions that determine who acquires sufficient legibility to enter the system’s evaluative space in the first place.