Active Carbon Accounting: Bayesian Value-of-Information Methods for Hardware Emissions Inventories

Reducing embodied greenhouse gas emissions from computing hardware requires inventories that identify the highest-emitting products and suppliers. Because product carbon footprints (PCFs) are scarce, organizations also use emissions estimates based on physical product attributes, averages for similar products, or purchase price. We propose active carbon accounting, a Bayesian framework that estimates emissions-estimation error by method and product category, propagates component-emissions uncertainty through hardware bills of materials (BOMs), and ranks missing product PCFs by their expected improvement to hardware-product emissions rankings per collection cost. The model treats PCFs as noisy emissions observations and represents repeated component occurrences and dependence among products from the same supplier. Preliminary paired-product and rack-level analyses show that the emissions-estimation method can materially change hardware inventory totals and that component-emissions uncertainty concentrates in components with large emissions contributions that recur within hardware~BOMs.