Carbon Footprint Context Cards: A Metadata Framework for Software Carbon Accounting

Abstract

While carbon-focused computing has emerged as a key approach for reducing the environmental footprint of cyberinfrastructure, widely used metrics that simultaneously incorporate operational and embodied impacts such as the Software Carbon Intensity (SCI) struggle to capture the behavior of modern, multi-layered, and multiplexed systems. We find that SCI-type metrics are underspecified, requiring many implicit assumptions to be made, such as how the embodied emissions of unallocated resources are attributed to applications, and the elasticity of the underlying computing infrastructure. Due to these implicit assumptions in current carbon accounting methodologies, application-level optimizations do not necessarily induce proportional—or even monotonic—reductions in system-level or global carbon emissions, revealing a mismatch and lack of causal alignment between local actions and global outcomes. To address the inconsistencies caused by the loose specification of current metrics, we propose Carbon Footprint Context Cards (CFCCs) to explicitly document key measurement assumptions. These cards aim to improve the specificity and interpretability of footprints by capturing essential metadata—such as embodied attribution methods, system boundaries, and resource elasticity—thereby guiding more effective and causally meaningful carbon optimizations.