Measuring the Unmeasurable: A Systematic Review of AI Environmental Footprint Methods and the SAERP Framework for Standardized Reporting

This deposit contains the PDF of the pre-print and the supplementary material for the systematic review "Measuring the Unmeasurable: A Systematic Review of AI Environmental Footprint Methods and the SAERP Framework for Standardized Reporting". It contains the complete empirical corpus of 174 studies (66 extracted variables plus source-filename traceability), the classified and venue-enriched versions used for the paper's statistics, the exclusion log for all 136 excluded papers, the raw 306-paper extraction output with per-paper confidence and verification-turn diffs, the reproducible LLM-assisted extraction pipeline (GLM 5.3 Flash via OpenRouter, two-turn verification), a deterministic classifier, and the analysis script that regenerates every reported statistic. It also contains the Standardized AI Environmental Reporting Protocol (SAERP) as a three-tier reporting framework (Bronze / Silver / Gold) with a JSON schema, a CSV template, and a print-ready checklist. Headline findings from the paper: 89.1% of studies report energy but only 55.7% report carbon, 19.5% embodied carbon, 6.3% water, and 2.9% rare-earth; Bronze aggregate compliance is 42.5% (74/174) under the strict rule; all 18 industry-funded studies were rated Medium or High Risk of Bias versus 26.9% of non-industry studies (Cramer's V = 0.468).