95% conformal prediction interval (global, n=21 calibration): 291,976 – 997,009 tCO₂. Restricted to steel-EAF stratum: 521,065 – 558,668 tCO₂ (subject to small-n undercoverage).
Single-site EAF plant — direct disclosure
The headline method is a closed-form physics baseline: Scope 1 ≈ nameplate capacity × route emission-factor × capacity factor, using only operator-published numbers. It is evaluated leave-one-plant-out: each plant's route emission-factor is derived only from the other audit-grade plants in its route, so no plant contributes to its own prediction. Headline: +82.3% log-MAE reduction vs the EU CBAM default across the 19 validatable plants (of 21 audit-grade; two single-plant routes — N₂O-controlled and blender fertilizer — cannot be independently validated and are excluded). External validity: the same formula on 372 EUTL-verified EU cement installations gives a median ratio ≈1.0 vs the EU default's ≈2.5×. An optional in-browser WebGPU model (iz) is shipped as a demo only; it does not beat this formula at current data scale.
TR-MRV-Bench v0 is Apache-2.0 licensed. The benchmark, methodology, and per-facility data are public at
github.com/abgnydn/iz and iz-mrv.pages.dev.
Cite: Günaydın (2026), "TR-MRV-Bench: a public per-facility emissions benchmark for Turkish CBAM-scope
industry, with a physics baseline that beats the EU CBAM default by +82.3% (leave-one-plant-out, n=19)."