iz · tr-mrv-bench · policy submission
Position paper · Apache-2.0 · replicable For the December 2027 default-values revision
The EU's 2026 default emission values for several Turkish CBAM-scope goods — most clearly Portland cement, Scrap-EAF steel, and N₂O-controlled nitric acid — overshoot audit-grade truth by 2× to ~30×, depending on route. We submit empirical counter-defaults derived from 21 audit-grade Turkish facilities, 789 EU-verified facilities under the EU ETS (10,691 audited plant-years from the EU Transaction Log), asking the Commission to move away from sector-aggregate defaults toward per-route audit-anchored defaults with operator opt-in.
The 2026 CBAM definitive phase applies country- and product-specific default emission values to imports for which operators do not submit verified actual data. The current implementing regulation derives these defaults from the 90th percentile of global emissions intensity per route and adds a markup escalating to 30% by 2028 — a conservative anchor designed to push operators toward submitting their own data. We do not contest the markup logic. We do contest the empirical anchor.
On the four CBAM scopes relevant to Turkey, the 90th-percentile anchor mis-prices in both directions against audit-grade truth:
Replace the 90th-percentile global anchor with a route-specific median of audit-grade operator-disclosed Scope 1 from the importing country (where available), retaining the existing markup schedule but applied to the empirically defensible anchor instead of a worst-case stand-in. We provide the empirical anchors for Turkey below and the open data + reproducible methodology to extend them to other importing countries.
The policy direction we argue for is not new; the empirical infrastructure to act on it is. We position this paper explicitly against the existing record so the Commission can see exactly what it adds.
| Existing voice | What they established | What this paper adds |
|---|---|---|
| TÜRKÇİMENTO (public statements, Mar 2026; "Actual Emissions Must Be Taken into Account") | Rhetorical / advocacy case that Turkey's national MRV data (cement clinker ≈ 0.88 t/t) should replace the "Other Countries" default (1.551 t/t); that CBAM risks becoming a de-facto trade barrier. | The per-facility audit-grade dataset and reproducible formula that turns that advocacy into a verifiable empirical artifact — 21 plants with provenance tags, 10,691 EUTL plant-years of cross-validation, open under Apache-2.0. |
| Bipartisan Policy Center (US) + the PROVE IT Act | The US needs its own emissions data because EU defaults are inconsistent with US ITC estimates; legislative push to document US emission-efficiency. | A working template for exactly that data effort, already executed for one non-EU country end-to-end. The methodology is country-agnostic and re-runnable. |
| Cembureau / Eurofer / BusinessEurope consultation responses | Industry positions on free allocation, scope, simplification, and administrative burden. | A per-route, per-CN-code default-value proposal backed by audited operator numbers rather than sector-level advocacy. |
| Solvay and the 246 Initiative-14828 Call-for-Evidence submissions (closed 25 Sep 2025) | Direct-vs-indirect scope, value-chain bookkeeping, verifier-cost concerns — mostly qualitative or firm-specific. | The first open, cross-validated, per-facility benchmark for a non-EU CBAM-exposed country. No prior submission cross-validates against the EU's own EUTL data or ships executable code. |
In short: industry bodies have argued that national MRV data should anchor defaults. This paper supplies the audit-grade data, the cross-validation against the EU's own verified emissions, and the reproducible method — so the argument can be acted on rather than merely repeated. To our knowledge it is the first such artifact for any non-EU CBAM-exposed country.
The TR-MRV-Bench is an open dataset of 59 CBAM-scope Turkish facilities (34 cement, 16 steel, 6 fertilizer, 3 aluminum). Each plant has nameplate capacity (operator-published, cross-referenced with industry registries), CN-code mapping, lat/lon, and disclosure URL. Of those, 21 plants have audit-grade Scope 1 emissions sourced from operator Integrated Annual Reports and sustainability statements (GRI for pre-2024 disclosures; TSRS-compliant for CY 2024 onwards following the KGK mandate effective 1 January 2024) along with ISO 14064-1 audit reports issued by TÜRKAK-accredited verifiers (and, for the sustainability-statement audit assurance layer, KGK-accredited audit firms). Provenance is tagged on every label: direct, allocated (from group total via clinker share), or composite (multi-site total).
In leave-one-plant-out cross-validation — each plant's emission factor estimated from only the other plants in its route, so a plant never helps fit itself — the closed-form physics formula tCO₂ = capacity × route-EF × capacity-factor achieves a +82.3% log-MAE reduction vs the EU CBAM default across the n=19 validatable plants (of 21 audit-grade plants; 2 are the sole plant of their route and cannot be cross-validated — BAGFAŞ, N₂O-controlled fertilizer, and Gübretaş, a blender). The earlier in-sample figure of +85.7% is optimistic because it lets each plant contribute to its own route fit; the leave-one-plant-out +82.3% is the honest number. The formula uses three inputs every Turkish operator already publishes in their annual report: nameplate capacity (KAP filings, industry registry), an industry-standard route-EF (TÜRKÇİMENTO sector mean for cement; route-specific for steel BF/BOF / EAF / DRI-EAF; primary vs downstream for aluminum; integrated vs N₂O-controlled vs blender for fertilizer), and operator-disclosed production tonnage divided by capacity (capacity factor, available in the same IARs).
To answer the obvious "this is just calibrated for Turkey" critique, we extracted verified Scope 1 emissions from the EU Transaction Log for every EU ETS installation in the four CBAM scopes: 372 cement clinker installations (5,198 plant-years), 282 pig-iron-and-steel installations (4,040 plant-years), 71 primary + secondary aluminum installations (724 plant-years), and 64 ammonia + nitric-acid installations (729 plant-years). Each is third-party verified annually under EU ETS Article 15. Total: 10,691 audit-grade Scope 1 facility-years.
Applying the same TR-derived formula to EU plants with no parameter re-tuning:
| Sector | EU plants pulled | EUTL verified total (Mt/yr) | Formula sector aggregate (Mt/yr) | Ratio | EU CBAM default applied (Mt/yr) | Ratio |
|---|---|---|---|---|---|---|
| Cement (clinker) | 372 | 102.4 | 106.1 | 1.04× | 261.4 | 2.55× |
| Aluminium primary | 31 | 5.77 | 5.40 | 0.94× | 3.97 | 0.69× |
| Steel (all routes) | 282 | 97.4 | 230.0* | 2.36×* | 218.5 | 2.24× |
| Fertilizer (ammonia) | 29 | 17.1 | 30.0** | 1.76×** | 30.0 | 1.76× |
*Steel aggregate is route-mix-sensitive: EU has wide capacity-utilisation spread across BF/BOF and EAF lines; per-plant log-MAE drops 54% vs EU default on a hand-curated n=12 sample. The aluminium row uses the 31 primary smelters (of 71 EU aluminium installations); the fertilizer row uses the 29 ammonia plants (of 64). **Fertilizer needs EU-specific EF: EU SMR-ammonia averages 1.14 t/t (best-in-class) where TR integrated is 2.0 t/t — the formula structure is right; the route-EF needs country-specific tuning.
The conclusion is uncomfortable for both directions: the EU CBAM defaults are systematically wrong against EU's own audit-grade data, in directions that depend on sector. Cement defaults overshoot EU truth by 2.55× (the EU's own plants would be over-tariffed by default). Primary aluminium Scope-1 defaults undershoot truth by 31%. A consistent methodology for setting defaults would address both.
We propose the following audit-grade anchors for Turkey-specific defaults, derived from operator IARs and sustainability statements verified by TÜRKAK-accredited GHG verifiers under ISO 14064-1 (and, for sustainability-statement assurance, KGK-accredited audit firms under TSRS for CY 2024-onwards filings). The markup schedule (10/20/30% for non-fertilizer, 1% for fertilizer) is retained as-is; only the empirical anchor changes. EU-default columns reflect the values applied to Turkish imports under the current implementing regulation (IR 2025/2621) and its predecessor (transitional Annex IX), depending on whether a Turkey-specific value has been published — Turkey currently falls under "other countries" for several CN codes. Where the current EU default could not be verified to a specific Annex entry from publicly available sources, we leave the column blank and rely on the structural argument in §4.
| CBAM route | CN codes | Current EU 2026 default for TR (t/t) | Proposed TR audit-grade anchor (t/t) | Empirical basis | Verifier coverage |
|---|---|---|---|---|---|
| Cement clinker (Portland) | 2523 21 / 2523 29 | 1.551 (Turkey, IR 2025/2621); 1.584 ("other countries" transitional) | 0.643 | TÜRKÇİMENTO sector mean 2023; 8 plant-level audit-grade Scope 1 disclosures (Akçansa Büyükçekmece/Çanakkale/Ladik 2025, Bursa 2024, Nuh 2024, Afyon 2024, Batısöke 2024, Göltaş 2024) + 3 group-level (Çimsa 2023, OYAK 2023, Limak 2023) | ±15% on 5/8 plant-level + 2/3 group-level (in-sample fit); overall leave-one-plant-out log-MAE 0.206 across all 19 plants; B6 EU-cement validates the formula structure on 372 EU plants |
| Steel — BF/BOF | 7208 / 7210 / 7211 (flat); 7213/7214 (long, BF/BOF route) | EU 2026 benchmark 1.370 t/t (route-specific, free-allocation reference); Turkey default not separately verified from publicly available sources | 2.000 | Erdemir Ereğli 2024, İsdemir 2024, Kardemir 2023 audited Scope 1 | n=3 plants; per-plant ratios 1.00 / 0.97 / 0.87 (within ±13%); matches global industry-standard EF for integrated long-flat mills |
| Steel — Scrap-EAF | 7213 / 7214 / 7227 (most TR exports) | EU 2026 benchmark 0.072 t/t (best EU EAF, free-allocation reference); default for non-EU EAF imports is several-fold higher than the EU benchmark | 0.250 | Çolakoğlu Gebze 2024, Habaş Aliağa 2024, İzdemir Aliağa 2024 — 3 plant-level audit-grade EAFs; Tosyalı holding 2022 group-level disclosure additionally covers 2 EAF sites (İskenderun, Sivas) | n=3 audit-grade + 2 group-level; formula EF 0.250 t/t crude steel |
| Steel — DRI-EAF | 7213 / 7214 (DRI-fed) | EU 2026 benchmark 0.481 t/t; Turkey default not separately verified | 0.400 | Tosyalı Osmaniye 2024 (TR's only DRI-EAF mill) | Anchor only — small sample (n=1) |
| Aluminium — primary unwrought | 7601 10 / 7601 20 | EU benchmark 1.423 t/t Scope 1 (free-allocation reference); country defaults higher (China 3.0, India 1.87, UAE 1.87, Turkey not separately published) | ~2.140 (EU primary fleet avg) | EUTL verified across 31 EU primary smelters (5.77 Mt / 2.7 Mt production, latest year) | EUTL audit-grade; TR's only primary smelter (Eti Seydişehir) does not publish Scope 1 separately |
| Aluminium — downstream (rolling/extrusion) | 7604 / 7605 / 7606 | Current EU defaults for downstream goods are calibrated to embed both Scope 1 + Scope 2 (electricity); we did not verify a specific Scope-1-only downstream default from primary sources | 0.450 | Assan Tuzla 2024 (108,500 tCO₂ on 286,119 t production → 0.379 t/t product), ASAŞ Akyazı 2024 audited | n=2 audited TR downstream plants; formula uses 0.450 t/t as a conservative buffer above audited EF; ratios 1.19 / 1.39 (formula slightly over-predicts both) |
| Fertilizer — integrated (NH₃ + urea + nitric) | 2814 (NH₃) / 2808 (HNO₃) / 3102 / 3105 | 0.800 (transitional "other countries" for fertilizer); 2026 country-specific values not separately verified | 0.500 | Toros Mersin/Samsun/Ceyhan 2024 audited group total (842,174 tCO₂) allocated to 3 sites by capacity | n=3 allocated; per-site ratios 0.77/0.82/1.46; group-aggregate well-anchored, per-plant variance reflects allocation uncertainty |
| Fertilizer — N₂O-controlled nitric acid | 2808 (with N₂O catalyst on tail gas) | Current EU defaults do not differentiate N₂O-controlled vs uncontrolled nitric acid; integrated fertilizer default applies | 0.050 | BAGFAŞ Bandırma 2024 audited 9,828 tCO₂ on 700,000 t capacity → effective ~0.014 t/t product; bench EF 0.050 is a conservative buffer above audited | Single TR operator; technology is well-documented industry-wide (95% N₂O removal verified by multiple OEMs); bench formula over-predicts by ~78% on this operator (conservative direction) |
| Fertilizer — blender only (no NH₃ process) | 3105 | No separate blender default; integrated default applies | 0.025 | Gübretaş Yarımca 2024 audited (granulation only, no NH₃/HNO₃ process) | Single TR operator; process emissions are physically near-zero (no high-T calcination) |
On the per-plant ratios above: the ratios in the "Verifier coverage" column are the proposed anchor's in-sample fit — each anchor is a route median or sector mean of the very plants listed, so it reproduces them by construction. The out-of-sample generalization evidence is the +82.3% leave-one-plant-out reduction (n=19) and the B6 cross-check on 372 EU cement plants — not the in-sample per-plant ratios. All TR audit-grade anchors are sourced from operator IARs and sustainability statements (GRI pre-2024; TSRS-compliant CY 2024-onwards per KGK mandate) verified by TÜRKAK-accredited GHG verifiers (ISO 14064-1) and, where applicable, KGK-accredited audit firms. Specific source PDFs are cited per facility at iz-mrv.pages.dev/bench. Reproducible scripts: bin/lopo_ef_eval.py, bin/baselines.py, bin/verifier_b6_eutl_score_sector.py at github.com/abgnydn/iz. We have not independently verified every numeric value in the current implementing regulation's annexes; where a specific EU default or free-allocation benchmark for a CN-code is shown above, it is sourced from secondary references (Türkçimento, Eurometal, S&P Global, EY publications) and may have been superseded by updates published after this submission. The structural argument in §4 does not depend on the exact current default values.
The Commission's choice of a 90th-percentile global anchor was a reasonable starting position when verified per-country data did not exist. As of 2026 it does — for cement, EU ETS Article 15 verified data covers 90%+ of European production; for Turkey, plant-level audit-grade IARs cover ~42% of CBAM-relevant capacity directly, rising to ~68% when group-level disclosures (Çimsa, OYAK, Limak) are included as anchors; for ammonia and primary aluminium, EUTL covers the entire EU production. Persisting with the 90th-percentile anchor produces three pathological outcomes:
The fundamental issue is that "90th percentile of global emissions intensity" is calibrated to penalise data-shy operators by making the default unattractive. That goal is preserved by the 10/20/30% markup. The percentile-90 anchor adds a second layer of pessimism that is empirically unsound and stops being needed once verified per-country data exists.
Should the Commission find this submission useful, we commit to maintain TR-MRV-Bench under Apache-2.0 with annual refresh of audit-grade labels each April (following the Turkish IAR publication cycle), to provide the same per-route audit-grade extraction for any other importing country whose operator IARs we can crawl (we currently have partial coverage of Greece, Romania, Spain via Holcim/Titan/Cementos Argos disclosures), and to keep the formula and the leave-one-plant-out methodology in the public domain. All code, all data, all formula derivations are at github.com/abgnydn/iz under Apache-2.0.
Respectfully submitted to the European Commission's CBAM unit (DG CLIMA + DG TAXUD) as a position paper for the mandatory default-values revision due by December 2027 under Regulation (EU) 2023/956 Article 7(7) and Implementing Regulation (EU) 2025/2621. Also intended for the Draft Act feedback phase of Initiative 14828 when that phase opens.
Ahmet Barış Günaydın · TR-MRV-Bench (iz) · hi@barisgunaydin.com · iz-mrv.pages.dev