ITEA — Measuring AI's impact at the right resolution: from occupations to activities, and why training must follow them v2.0

Replication workbook and manuscript for the ITEA Framework. The study shows that most AI-exposure indices share a unit-of-measurement error —the occupation— and that 82–94% of the variance in task-level AI exposure and knowledge codifiability lives *within* occupations. Measurement is reframed onto the work *activity* (O\*NET DWA), graded by the AXI apparatus across three technologies (classic automation, generative AI, agentic AI), codifiability/expropriability and responsibility, and validated in its own unit against real AI use (Anthropic Economic Index; 88.8% match; Spearman +0.17 with GenAI, ≈0 with classic automation). The workbook (XLSX, 12 sheets) contains the 47,810 graded tasks, the AEI real-use table (17,426 tasks), the occupation master (784), the variance decomposition (Table 1), the incremental-validity regression (Table 2), the robustness battery, and the source tables for Figures 6–7.- **Version:** `v1.8`