Mona Baraké, Theresa Neef, Paul-Emmanuel Chouc, Gabriel Zucman · EU Tax Observatory Note, July 2021

Minimizing the Minimum Tax? The Critical Effect of Substance Carve-Outs

13 problems found: 4 moderate · 9 minor

What the paper claims

A 5% substance-based carve-out reduces EU revenue from a 15% global minimum tax by about 15%.
A 5% carve-out reduces EU revenue from a 25% minimum tax by about 21%.
The transitional 7.5% carve-out (and a 10% carve-out) cut EU revenue further.
Overall, carve-outs reduce EU minimum-tax revenue by 15% to 30%, depending on the carve-out and minimum rates.
Carve-outs remove a much larger share of domestic and non-haven profits than of tax-haven profits, so the revenue cut is largest for domestic and non-haven revenue.
The higher the minimum rate, the more carve-outs weigh on revenue, because a low minimum rate draws revenue mainly from havens.
Composition and OECD comparison: payroll is 11% and tangible assets 89% of the carve-out; aligning with the OECD (2020) definition gives a reduction of slightly less than 4%, versus about 3% in the OECD's Impact Assessment.
Conceptually, carve-outs make any tax rate (even 0%) acceptable for profits matched by substance and encourage moving real activity to low-tax places.

Every problem found

1moderateconfirmed · newly identified here · methodology

For 13 EU countries without detailed multinational data, the note uses the profits of all domestically owned companies, including small, purely domestic firms that the minimum tax would never cover. This inflates the figures for countries such as Poland and Germany and slightly enlarges the estimated carve-out loss. Even removing this component entirely still leaves a loss of 13-17%.

Technical detail and evidence

Why it matters, in detail. The domestic component carries the highest haircut (33%), so overweighting it inflates both the EUR levels and, modestly, the percentage loss. The EUR effect is concentrated at 25%, where TWZ local-firm ETRs of 17-23% bind. Rescaling toward in-scope MNEs (ratio about 0.35-0.4 for DEU/ESP) lowers the 25% loss from 21.1% to roughly 19-20% and the 25% EU baseline by roughly EUR 15-25 bn. At 15% the effect runs mainly through Poland (all-corporation ETR 10.0%). The sign is not fully determined, because in-scope groups' domestic ETRs may differ from all-corporation ETRs.

Evidence. Note p.7. Companion report (July 2021 revision) Appendix A, pp.36-37: 'profits of local firms and their ETRs provided in Tørsløv, Wier, Zucman (2018, Appendix Tables, Table A6)'. Inputs (twz_domestic.csv): DEU $522.9 bn (ETR 22.75%), ESP $141.6 bn (17.9%), POL $70.1 bn (10.0%). In the first year with CbCR data (2017), in-scope domestic profits are 39% (DEU), 34% (ESP) and 11% (GRC) of these TWZ figures. Adjudicator decomposition (outputs/sensitivity_info.json): at 25% the imputed domestic components contribute EUR 39.3 bn of the 167.7 bn baseline and 13.1 bn (37%) of the 35.4 bn loss. At 15%, Poland's proxy (EUR 3.6 bn of Poland's 3.7 bn) produces 1.2 bn (16%) of the 7.3 bn EU loss. Bounds: halving these bases gives -14.3%/-19.5% (baselines 46.1/148.0 bn); removing them gives -13.4%/-17.4% (43.9/128.4 bn); scaling DEU/ESP/GRC by the 2017 ratios gives -15.2%/-19.9% (152.3 bn at 25%). Auditor B's figures match exactly.

Why this severity. Even the extreme bound (no domestic base at all for the imputed countries) leaves EU losses of 13.4% and 17.4%. Realistic in-scope shares move the 25% headline by about 1-3 pp and its EUR baseline by roughly 10-20%: noticeable, not transformative.

2moderatelikely · previously known · methodology

The note counts goods held in stock (inventories) as 'substance' that shields profits from the minimum tax, but the rules eventually adopted do not count them. Leaving them out shrinks the estimated revenue loss by about a fifth, from about 15% to 12% at a 15% minimum rate. The loss is still large, just smaller.

Technical detail and evidence

Why it matters, in detail. Tangible assets supply 89% of the carve-out, so excluding inventories (OECD's BEA-based 24% average) scales the asset term by 0.76. The EU loss from a 5% carve-out falls from 15.2% to 12.0% at a 15% rate and from 21.1% to 16.6% at 25%, so the summary range '15-30%' becomes roughly 12-25%. Inventory shares for non-US MNEs are unmeasured, so the true correction could be somewhat smaller or larger. Partial offset: under-measured payroll (no employer contributions; a 50% premium gives -15.6%/-21.7% before the inventory correction).

Evidence. Note pp.6, 13. Blueprint (Oct 2020) p.93 box, paras 342-347. Baraké, Chouc, Neef & Zucman (2022), Intertax 50(10), p.692 and Online Appendix 1 §B.6. Adjudicator re-run (scripts/sensitivity.py, S1: tangible assets x0.76): EU loss -15.2% -> -12.0% at 15% and -21.1% -> -16.6% at 25%; with a 7.5% carve-out, -22.6% -> -17.9% and -31.3% -> -24.9%. The overall carved-out share of positive CbCR profit falls from 27.5% to 21.6% (outputs/text_checks.json). The OECD's firm-level-adjusted model gives 20.5% for the long-run 5%/5% SBIE on PP&E (OECD Taxation WP 68, Hugger et al. 2024, para 80). Both auditors obtained the same figures (A: -12.0/-16.6; B: 11.99/16.65).

Why this severity. It lowers every headline percentage by about one-fifth: 3.2 pp at 15% and 4.5 pp at 25%. That is below the roughly one-quarter threshold for 'major'. The qualitative conclusion is unchanged, and the choice was disclosed and arguable under the July 2021 wording.

The authors' response. Intertax (2022, p.692): 'No adjustment has been made, but further guidance on this aspect might make a correction necessary. In that case, the results displayed below would overestimate the effect of the substance-based income exclusion.' Online Appendix 1 §B.6: 'the interpretation of the model rules points to an exclusion of inventories'. With the 24% haircut their 2017-data EU reductions fall from 18% to 14% (long-run rates) and from 29% to 24% (year-one rates).

3moderatelikely · previously known · data error

A few suspicious entries in the 2016 data, such as one Belgian entry showing almost 90 cents of profit per euro of sales, inflate the euro amounts the note reports. Fixing them lowers the euro figures by up to about a fifth, but makes the carve-outs' percentage impact look even larger, so the note's main point is not weakened.

Technical detail and evidence

Why it matters, in detail. Both anomalies inflate low-taxed profit with little substance attached (haven-type cell; dividend-inflated domestic profit), which raises the no-carve baseline more than the carve-out. Correcting them lowers the EUR 48 bn / 168 bn baselines to roughly EUR 38-44 bn / 135-160 bn and raises the percentage losses to 17-20% (15%) and 22-28% (25%). The percentage headline is therefore conservative in this respect; the EUR levels are fragile. On 2017 data the authors' own EU baseline at 15% is EUR 83 bn.

Evidence. oecd.csv cell BEL/NLD 2016 (positive-profit panel). Authors' EU Tax Observatory Note 2 (Oct 2021), Appendix C (Tables A7-A9; 2017 margin 29.6%) and Appendix B (Table A5; factor 0.6028). Adjudicator re-run: Belgian margin fix (ETR fixed) gives EU 43.65 -> 36.24 bn at 15% (-17.0%) and 159.8 -> 124.2 bn at 25% (-22.3%). The dividend factor on CbCR domestic cells gives 42.65 -> 35.16 bn (-17.5%) and 142.65 -> 104.78 bn (-26.5%). Both together give 38.0 -> 30.5 bn (-19.9%) and 134.8 -> 96.7 bn (-28.2%). The auditors' figures match.

Why this severity. The corrections cut the summary's EUR levels by about 10-21% at 15% and 5-20% at 25%. They raise rather than lower the percentage reductions, so they qualify the levels without undermining the percentage finding.

The authors' response. Note 2 (Oct 2021), Appendix C: 'In the absence of any explicit correction ... we base our benchmark estimates on the unadjusted country-by-country data. This may lead to overstating the Belgian revenue gains'. Appendix B proposes the 0.6028 rule-of-thumb factor (EU 15% revenue about EUR 80 bn -> 67 bn on 2017 data). Neither adjustment was applied to the July note.

4moderateplausible · newly identified here · speculative assumption

About half of the EU-wide loss comes from countries whose carve-out effect is not measured directly but borrowed from averages over other countries, an average heavily influenced by Chinese companies with huge domestic asset holdings. Using European averages instead changes the result by a couple of percentage points, in either direction depending on the choice, so the headline is less precise than its decimals suggest.

Technical detail and evidence

Why it matters, in detail. The 13 imputed countries account for 39-50% of the EU loss and their percentages are set entirely by transferred donor ratios. The donor set (global, EU, ex-China, ex-USA) spans -13.6% to -16.9% at 15% and -18.3% to -23.7% at 25%, against -15.2%/-21.1% published. China's asset-heavy domestic profile raises the domestic donor share by about 7 pp, while dividend double counting in CbCR domestic profits lowers it, so the net bias is unsigned.

Evidence. Note p.7. Adjudicator re-run (scripts/sensitivity.py S6-S11; outputs/sensitivity_info.json): EU-parent domestic share -> -14.4%/-18.9%; all donor shares from EU parents -> -13.6%/-18.3%; donors excluding China -> -14.2%/-19.2%; donors excluding USA -> -16.9%/-23.7%; low-taxed donor cells only -> -14.2%/-20.9%. The 2017 domestic share (28/21/10% for domestic/non-haven/haven) is from Intertax 2022 Online Appendix 1 §B.2. In the authors' Note 2 (Oct 2021, 2017 CbCR, Table 3), Germany (-24%) and Spain (-31%) show larger 15%-rate reductions than in the July imputation (-16% each), so there is no evidence of upward bias at 15%. Double-counted dividends (F4) push the true domestic share the other way.

Why this severity. Reasonable donor choices move the EU headline by about +/-1.5-2 pp at 15% and +/-2.5-3 pp at 25%, in both directions. This is a robustness and disclosure concern, not a demonstrated bias.

5minorconfirmed · newly identified here · internal inconsistency

Part of what the note calls 'non-haven' income is really unidentified income, some of it taxed almost nothing and probably sitting in tax havens. Sorting it out would make the note's point, that carve-outs mostly protect income where firms have real operations, even stronger.

Technical detail and evidence

Why it matters, in detail. More than half of the 'non-haven' column is mixed or residual geography, some of it taxed at havens' rates. Separating identified non-havens raises their revenue cut from 16% to 22%, and adding residual rows to havens leaves havens near 7.5%. The published 23%/8% carved-out shares mix the two groups.

Evidence. Simulator calculator.py lines 965-975 (aggregates assigned to non-haven); adjudicator scripts/text_checks.py: identified non-haven revenue 15.96 -> 12.39 bn (-22.3%) vs labelled -16.1%; haven plus residual rows -7.5%. Non-haven carved-out share is 25.4% on identified partners (23.0% published); the haven share becomes 11.5% if residuals are counted as havens. Auditor B's narrower figures (7.447886/6.611174 bn; -17.35%) match.

Why this severity. Totals are unchanged, and reclassifying the residual rows sharpens rather than weakens the haven/non-haven contrast in C5. This is a labelling problem.

6minorconfirmed · newly identified here · internal inconsistency

A technical quirk inherited from the authors' earlier report makes some country figures at a 21% minimum rate look odd (smaller effects than at 15%). It does not affect the EU-wide numbers or the note's general point that higher minimum rates make carve-outs more costly.

Technical detail and evidence

Why it matters, in detail. The switch between imputation regimes changes the latent non-haven base between 20% and 21%, so the 21% column mixes base assumptions. Repairing continuity restores monotonicity for the imputed countries and barely moves the aggregates (EU -18.4% -> -18.7%; full sample -17.8% -> -17.9%). Remaining non-monotonicities (e.g., ZAF, CHL, IND) reflect the data.

Evidence. Companion report Appendix A, eq. (A2) and following paragraph; calculator.py lines 542-671 and 905-948 (read as text); adjudicator rate grid and continuity repair (outputs/text_checks.json): EU 21% revenue pair 106.1/86.3 bn and -18.7%, against the printed -18.4%. Auditor B's numbers match (94.568/84.090; 106.124/86.296).

Why this severity. Only the country-level 21% values in Table A3, a secondary table the text does not discuss, are distorted. The aggregate 21% figure moves by 0.3 pp and C6 holds in aggregate.

7minorconfirmed · newly identified here · documentation gap

When comparing its result with an earlier OECD estimate, the note leaves out that the OECD excluded US companies. Like-for-like, the two estimates are a bit further apart (about 5% vs 3%), but the explanation the note gives for the gap still stands.

Technical detail and evidence

Why it matters, in detail. On the OECD's own scope the aligned estimate is about 5%, not just under 4%, so the note's estimates sit further from the OECD's than presented. The main explanation (depreciation-only asset term and inventory exclusion versus full carrying value) still holds.

Evidence. Note pp.12-13. OECD (2020) Economic Impact Assessment, Table 3.6 title and notes, para 47. Adjudicator: -3.68% including US parents and -5.06% excluding them; payroll share 61.1% (outputs/text_checks.json). Matches both auditors.

Why this severity. It affects a side comparison only. Definitions still explain most of the gap between about 3% and 15-21%.

8minorconfirmed · newly identified here · documentation gap

Much of the estimated loss comes from profits multinationals make in their home country, which the note assumes home countries will also tax at the minimum rate. That is a reasonable assumption for EU countries, but the note does not say it, and without it the loss would be about a quarter smaller.

Technical detail and evidence

Why it matters, in detail. Excluding domestic top-up lowers the EU loss to 11.2%/14.8%, because domestic profits carry the largest carve-out share (33%). The headline is therefore conditional on HQ countries taxing domestic low-taxed profit, which is plausible for EU states.

Evidence. July 2021 statement, 'Rule design'; companion report (June 2021) p.11; adjudicator S21: foreign profits only gives -11.2% at 15% and -14.8% at 25% (outputs/sensitivity.csv). Intertax (2022, p.692) bases the domestic top-up on the European Commission's directive proposal (the adopted Directive 2022/2523 text was not retrieved here).

Why this severity. The assumption is disclosed in the companion report, consistent with the later EU approach and with domestic top-up taxes, and Table A2 shows the domestic component separately.

9minorconfirmed · newly identified here · overclaiming

The note suggests that with carve-outs any tax rate, even zero, becomes acceptable. That is true only for a modest 'normal' return on real assets and workers; profits above that are still taxed.

Technical detail and evidence

Why it matters, in detail. The claim generalises from the case where profit equals 5% of substance. In general the exclusion is a fixed routine return and the residual remains subject to top-up. The relocation incentive is real but its size is not estimated.

Evidence. Note pp.3-4 and shares on pp.5, 11. Blueprint para 333 acknowledges that an MNE 'can increase the amount of the carve-out by shifting more payroll and tangible assets into the jurisdiction', so the incentive argument itself is sound.

Why this severity. This is framing only; the quantitative findings and the marginal-incentive argument are unaffected.

10minorconfirmed · newly identified here · documentation gap

The note estimates wage bills from national average wages, with some countries assigned regional averages. This is imprecise and not fully documented, but it barely affects the results.

Technical detail and evidence

Why it matters, in detail. Payroll is a small, noisy component. Wage imputations and premium choices move EU losses within 14.9-15.6% (15%) and 20.7-21.7% (25%).

Evidence. Note pp.5-6; preprocessed_mean_wages.csv; adjudicator text_checks.py (95 own values covering 76.6% of positive profit, consistent with the note's '>75%'). The payroll premium at 0% or 50% gives -14.9% to -15.6% at 15%. Replacing five conspicuous imputations with the ILOSTAT-based values reported by Auditor A (NLD 53k, DNK 59k, MEX 3.7k, IND 2.8k, BMU 75k USD) gives -15.21%/-21.14%.

Why this severity. Payroll is only 11% of the carve-out, and every tested correction moves the headline by under 0.5 pp.

11minorconfirmed · newly identified here · internal inconsistency

Data the authors said they excluded still slip into one intermediate average. The effect on results is negligible.

Technical detail and evidence

Why it matters, in detail. A small inconsistency between the stated sample and the donor sample; immaterial.

Evidence. Note p.11 fn 1; calculator.py lines 429-456 and 886-887 (read as text). Adjudicator S12: domestic share 33.30% -> 33.66%; EU post-carve revenue 40.936 -> 40.921 bn (15%) and 132.268 -> 132.129 bn (25%); losses -15.25%/-21.20%. Matches Auditor B.

Why this severity. The effect is about 0.03-0.08 pp and slightly increases the loss.

12minorconfirmed · newly identified here · documentation gap

For some countries the note uses estimates of profits 'shifted' to tax havens but describes them as all profits booked there. The mix-up matters little for the percentage loss (under one point), but readers should know what the numbers measure.

Technical detail and evidence

Why it matters, in detail. Applying a substance haircut to a shifted-profit base slightly overstates the carve-out's bite. Using total booked haven profit instead would raise the haven weight (low 7.7% share) and also lower the EU percentage. Either way the published percentage is a weak upper bound in this respect (about 1 pp). Corpus critiques that TWZ shifted profits are too high (offshore-havens-09, -11, -12) would lower EUR levels but raise the percentage loss (TWZ haven profits x0.5: -15.5%/-21.9%).

Evidence. twz.csv equals TWZ2019_2016update.xlsx, sheet 'TableC4', columns [12]-[22] (the 'Ultimate ownership' block, e.g. Germany 'All havens' 38,561.2 $m); the simulator data README says 'amounts of profits shifted to tax havens ... re-allocated ... on an ultimate ownership basis'. These inputs supply EUR 6.8 bn (14%) of the EU baseline at 15% and 20.3 bn (12%) at 25%. If no carve-out were applied to them, the EU loss would be about EUR 0.5 bn smaller at 15% (-14.1% instead of -15.2%) and 1.6 bn smaller at 25% (-20.2% instead of -21.1%).

Why this severity. It moves the percentage headline by at most about 1 pp; otherwise it is a labelling and levels issue inherited from the companion report.

13minorlikely · previously known · methodology

The note pools many companies together, which lets one firm's unused allowance shelter another firm's profit. Evidence from company-level data shows this matters little: it changes the result by less than one percentage point.

Technical detail and evidence

Why it matters, in detail. The bias is upward by construction for the cap, bounded by OECD estimates of 'SBIE usage' of 0.95-1.00 at long-run rates. Interaction with within-cell ETR heterogeneity is unsigned; Auditor A's Table IV test suggests it does not inflate the effect.

Evidence. Note p.6; simulator calculator.py lines 405-418 (read as text). In the 2016 cells the cap binds for only 0.13% of profit (outputs/text_checks.json). OECD Taxation WP 68 (Hugger et al. 2024) addresses exactly this aggregation (para 78 and Annex C.2): using Orbis subgroups, 95-100% of substance is usable at 5%/5% rates and 90-100% at 8%/10%. Auditor A's test with OECD Table IV ETR bins found bin-level carve-out effects at least as large as cell-level ones, so ETR heterogeneity does not reveal an exaggeration either.

Why this severity. External firm-level evidence implies that pooling overstates the carve-out by at most about 5%: no more than about 0.8 pp on the 15.2% headline and 1.1 pp on 21.1%.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

I4: Brazil's -23.2% at a 21% rate in Table A3 is a transcription slip (copied from Australia's row); the full-sample 21% entry (-17.8% vs -17.73%) is a rounding difference. — Both printed cells are reproduced exactly when the list of countries whose TWZ haven deficit replaces the CbCR one is set from the no-carve-out run at the same minimum rate, rather than fixed at 25% as in the archived code (b3b0e47, committed 23 July 2021, after the note was online on 21 July). Under that rate-consistent rule all 967 printed cells match (adjudicator scripts/reproduce_tables.py; Auditor B found the same); Brazil is selected at 25% but not at 21%. -17.73% would round to -17.7%, so 'rounding' does not explain the full-sample cell either. The equality with Australia's -23.2% is a coincidence. What remains is a trivial provenance point: the archived simulator code does not regenerate two printed cells.
Checks passed #7: the simulator quirk that fixes the TWZ-for-CbCR replacement set at 25% 'changes no published number'. — The quirk is precisely what produces the two Table A3 discrepancies; the printed numbers correspond to the rate-consistent rule.
I2 evidence: 'the authors' own later estimate, on later data, is 22%' for the domestic carved-out share. — Misread. 22% (Intertax 2022, Online Appendix 1 §B.6) is the share after excluding inventories; the like-for-like figure with inventories on 2017 data is 28% (§B.2). The direction of Auditor A's point (33% is high) survives, but the gap is 5 pp, not 11.
I3 range: applying the 0.6028 dividend factor to TWZ domestic profits as well (with TWZ ETRs rescaled) gives -13.6%/-18.2% (S10). — Not a valid correction. TWZ local-firm profits come from national accounts, which exclude intra-group dividends, so scaling them (and inflating their ETRs, e.g. Germany to ~38%) has no basis. The scenario is not used to bound the headline.
Inventory inclusion is an explicitly reasoned policy-definition choice; excluding inventories is 'a different policy-definition scenario', not an issue. — The Blueprint on which the July statement built, the December 2021 Model Rules (Art. 5.3.4) and the authors' own 2022 paper all point to excluding inventories. The choice is disclosed, but it biases the headline upward by about one-fifth, so it is graded as an issue (F1).
I2: aggregate caps and donor transfers form a moderate, 'confirmed' limitation whose empirical bias cannot be quantified. — Partly rejected on severity. The cap-pooling component is bounded by OECD WP 68 (2024) firm-level evidence (95-100% SBIE usage at 5%/5%), implying at most about 0.8 pp on the 15% headline, so it is graded minor (F5). The donor-transfer component is merged into F3.
I6: the 20% imputation discontinuity is a moderate issue. — Confirmed but downgraded to minor (F7): for this note it affects only the country-level 21% cells of Table A3; the EU 21% percentage moves by 0.3 pp under a continuity repair, and the 15%/25% headlines are untouched.

Audit notes

What reproduced. Independent re-implementation (scripts/model.py, written from the note, the companion report's Appendix A and a text-only reading of the authors' simulator; no author code executed) on the authors' published inputs (tax-deficit-simulator @ b3b0e47 data files). Results: EU 15%: 48.282 -> 40.936 bn (-15.213%); EU 25%: 167.686 -> 132.268 bn (-21.121%); 7.5%: 37.372 bn (-22.6%) and 115.171 bn (-31.3%); 10%: 34.128 bn (-29.3%) and 100.316 bn (-40.2%). Full-sample and CbC totals also match (the printed 'CbC reporting countries' total excludes Sweden). Cell by cell across Tables 1, 2, A1, A2 and A3: 967/967 match with the rate-consistent TWZ-substitution rule; 965/967 with the archived code's fixed-25% rule (Brazil 21%: -21.3% vs printed -23.2%; full sample 21%: -17.73% vs -17.8%); 947/967 with the 12-July-2021 code's independent-maximum rule. In-text statistics reproduce: carved-out shares 33.3/23.0/7.7%; payroll 10.7% of the carve-out; CbC -12.3%; domestic/non-haven/haven components 21.27->16.44, 37.42->31.38, 56.07->52.77 bn; footnote 3 26.6% (n=526) and 5.6% (n=213); OECD-aligned -3.68% with payroll 61.1%. 25 spot-checked US input values (five partner rows x five variables) equal the IRS 2016 CbCR workbook (Auditor B: 400/400). The baselines equal the July-revised/HAL companion report (48.3/167.7). The three note versions (HAL v1; EUTO July 2021 PDF = GitHub copy, same SHA-256; Dec 2022 re-layout) contain identical sequences of 1,214 numbers. Not reproduced: the ILO wage preprocessing (unpublished; the '104 of 173' coverage count could not be matched, 95 of 174 found, though the >75% profit share holds) and the 2016-USD -> 2021-EUR factors, which were taken from the authors' files (both auditors re-derived them from ECB/IMF data).

Limits of this audit. No author code was executed; equivalence rests on an independent re-implementation that matches the printed numbers. The 2021 ILOSTAT vintage and the authors' wage preprocessing were unavailable; ILOSTAT comparison values for five partners are taken from Auditor A. The companion report's baseline inputs (TWZ data, haven list, 10% haven ETR, GDP uprating) were taken as given except where noted; they are audited separately. The 24% inventory factor is the OECD's BEA-based average and the inventory share of non-US MNEs is unmeasured. The in-scope share of TWZ local-firm profits is proxied only by 2017 CbCR data for Germany, Spain and Greece. No firm-level data were available to quantify the cap-aggregation bias, so the bound relies on OECD WP 68 (2024; PDF obtained from a Bocconi mirror). The adopted EU Directive 2022/2523 text could not be retrieved (EUR-Lex blocked automated access). All scenarios are static and mechanical, like the note itself; behavioural relocation of substance, which the note argues would raise the cost, is not modelled.