World Inequality Report 2022

Why it matters, in detail. The top end of the modelled carbon distribution is driven by the gamma = 1 wealth-based allocation of GFCF emissions (~32% of the total). That rule loads about 70% of the top 1%'s footprint onto wealth. Replacing it with a consumption-proportional allocation (close to capital endogenisation) or an income-proportional one lowers T1 from 16.8% to 8.2-11.4% and T10 from 47.6% to 37.3-41.7%, while B50 rises only from 12.0% to 12.6-13.9%. The top 1%'s share of 1990-2019 emissions growth falls from ~21% to 7-13%, below the bottom 50%'s 17-20%. The published sensitivity analysis (gamma 0.8-1.2, alpha 0.4-1) does not span these alternatives. The headline T1 and growth-attribution claims are therefore convention-dependent rather than empirically established.

Evidence. Adjudicator script carbon_allocation.py, run on the snapshot wid-data-25102021. - Published benchmark reproduced: aggregating the published country carbon distributions (lpfghg999i, 170 countries) gives 2019 B50 12.0%, T10 47.5%, T1 16.8%, T0.1 7.1%. Shares of 1990-2019 growth: T1 22.4% vs B50 16.7% (published 12.0/47.6/16.8/7.1 and 21% vs 16%). - Rule reproduced: rebuilding from footprint components (khfghg/kifghg/kgfghg/kofghg999i, alpha 0.6, gamma 1) gives 12.0/47.2/16.6/7.05. - Changing only the investment allocation: - like consumption: 2019 B50 13.9%, T10 37.3%, T1 8.2%, T0.1 1.9%; 1990-2019 growth T1 7.3% vs B50 19.6%; - proportional to income: T10 41.7%, T1 11.4%; growth T1 13.1% vs B50 17.3%; - wealth with elasticity 0.5: T10 39.0%, T1 8.9%; - investment emissions excluded: T10 38.1%, T1 8.4%. - Keeping the authors' own published consumption allocation and swapping only the investment term: T10 37.4%, T1 8.4%; 1995-2019 growth T1 6.1% vs B50 18.7%. - Auditor A obtained the same results with independent code (claude/scripts/carbon_allocation_sensitivity.py).

Why this severity. Under a standard, defensible alternative convention, three things change. The top-1% share roughly halves (16.8% -> 8-11%). The top-10% share falls by 6-10 pp (47.6% -> 37-42%). The headline claim that the top 1% added more to emissions growth than the bottom 50% reverses. The qualitative message that emissions are highly concentrated survives.

The authors' response. Chancel (2022; WIL WP 2021/21 and Nature Sustainability SI p.13) defends investor attribution: investors are the 'sole decision-makers' about investment, and the rule is 'transparent and plausible'. He cites estimates that the carbon content of wealth rises roughly one-for-one with wealth (about 1.1 in France and 0.95 in Germany), notes that saving is more concentrated than wealth, and calls gamma = 1 conservative. No results under consumer attribution are reported.

Why it matters, in detail. Using the published counts, wealth totals and marginal schedule, scenario-1 static revenue is $1.47-1.53tn. The report's own 10%/15% haircuts give 1.45-1.51% of 2021 world NI at MER (1.42-1.48% with the additive 25% haircut the published numbers imply), not 1.62%. Scenarios 2 and 3 recompute 7-9% lower (1.94% vs 2.08%; 4.81% vs 5.30%). The excess comes from first- and second-bracket effective rates (0.61% and 1.27%) that exceed the exact maxima implied by the bracket means ($1.84m and $19m). The generating file is missing, so the table cannot be reproduced. The printed schedule (2%) and the computation (1.5%) also disagree, which adds version ambiguity.

Evidence. Adjudicator check_wealth_tax.py, using WIR2022TablesFigures-Chapter7.xlsx (data-T7.1) and the snapshot denominator (WO-MER mnninc999i 2021 = EUR 65.403tn x 1.18585 = $77.56tn). - Exact static liability in scenario 1 is $1,526bn with the printed 2% schedule and $1,468bn with 1.5%. - After the report's 10% evasion and 15% depreciation (x0.765) this is 1.51% or 1.45% of world NI (1.48% or 1.42% with x0.75), against a published 1.62%. Scenario 2: 1.94% vs 2.08%. Scenario 3: 4.81% vs 5.30%. - Scenario-1 bracket static rates (printed schedule) are 0.457/1.184/1.750/2.311/2.788/3.121%, against published 0.61/1.27/1.48/2.25/2.83/3.20%. - Both auditors obtained the same figures (Codex 1.506/1.941/4.810%; A 1.51%).

Why this severity. Recomputed headline revenue is about 7-12% below the published figure (1.42-1.51% vs 1.62% of world income), and several bracket rates are impossible. The order of magnitude of the policy claim, that modest taxes on multimillionaires raise about 1.5% of world income, survives.

Why it matters, in detail. A fractile with a constant share s captures s of the increment, and the 38% equals the top-1% share in both 1995 and 2021. Over 1995-2021 the global top-1% wealth share was flat and the top-10% share fell, at both PPP and MER, while the bottom-50% share rose by about 26% in relative terms. Concentration rose only within the top 1%: top 0.1% +3.5 pp, top 0.01% +4.0 pp, billionaires 1.1% -> 3.3-3.5%. 'Serious inequality in growth rates between top and bottom' holds for the top 0.01% and above (5-9% a year) against the average. It does not hold for the top 1% (3.2%) against the bottom 50% (4.2%).

Evidence. Adjudicator wealth_growth.py (WO shweal992j and ahweal992j, 1995 and 2021). - Shares 1995 -> 2021: T1 37.72 -> 37.79%; T10 79.16 -> 75.56%; M40 19.26 -> 22.45%; B50 1.58 -> 1.99%; T0.1 15.91 -> 19.38%; T0.01 7.18 -> 11.20%. - Shares of the wealth increment: T1 37.8%; B50 2.1% (total basis) or 2.3% (per adult). - Annual growth (CAGR): mean 3.23%, B50 4.17%, T10 3.05%, T1 3.24%. - At MER the pattern is the same: T10 86.4 -> 82.4%, T1 42.8 -> 43.1%. - External: the Credit Suisse Global Wealth Report 2021 (p.24) shows the global top-10% share falling from 88.7% (2000) to 80.9% (2019) and the top-1% share from 48.3% to 43.8%.

Why this severity. No number is wrong. But a headline summary statement on the dynamics of global wealth is framed so that readers infer rising top-1% concentration, and it wrongly attributes the contrast to growth-rate differences that the data do not show for the top 1% vs the bottom 50%. The narrower claim of rising concentration within the top 0.1% is supported.

The authors' response. None located. The Foreword (p.3) acknowledges that the top 10% grew slower than the world average, and the Fig.9 note says the bottom 50% grew 3-4% a year from a very low base.

Why it matters, in detail. Household net wealth (the total of the WID wealth distributions) is below 50% of 'private wealth' (households plus NPISH by definition) in 95 countries with about 25% of world adults. Either Ch.3's private-wealth ratios for emerging economies are overstated, or household wealth, and hence those countries' position in the global distribution, is understated. Scaling household wealth up to 60-95% of private wealth where it is lower, shapes held fixed, cuts the global top-10% wealth share from 75.4% to 73.9-72.5% and raises the bottom-50% share from 1.9% to 2.4-2.6%. Which aggregate is wrong cannot be determined from the release.

Evidence. Adjudicator snapshot reads (2020: 145 of 174 countries have a household/private ratio below 0.9, holding 63% of world adults; 95 are below 0.5, holding 25%) and household_private_sensitivity.py. Scaling each country's household wealth (shape preserved) up to 95% of private wealth where lower: global T10 75.4% -> 72.5%, B50 1.93% -> 2.64%, T1 37.7% -> 35.9%. Scaling to 60%: T10 73.9%, B50 2.35%. Replicates Auditor A.

Why this severity. A reconciliation within the range implied by the report's own definition lowers the global top-10% wealth share by 1.5-3 pp and raises the bottom-50% share by a fifth to a third. The qualitative finding that the top 10% own about three-quarters survives.

Why it matters, in detail. Distribution-neutral nowcasting is applied from each country's last observed year. About 46% of 2021 adults live in countries whose income distribution was last updated in 2011-2015, and no country's distribution changes in 2021. The within-country T10/B50 plateau at 15.1 (2015-2020), which Ch.2 interprets substantively, is therefore largely mechanical for these countries. Part of the 2015-2020 fall in the global T10/B50 (40.3 -> 38.4) rests on frozen within-country shapes. The level risk looks small: the current WID puts 2021 global T10 at 53.4% and T1 at 20.2%, against the report's 52.2% and 19.3%.

Evidence. Adjudicator distribution_vintage.py (sptinc992j shares, 1995-2021), giving the last year in which each country's distribution changes, weighted by 2021 adults: - 2010 or earlier: 0.4% - 2011-2015: 46.5% - 2016-2017: 9.6% - 2018: 16.2% - 2019: 17.9% - 2020: 9.4% - 2021: 0% Examples: China 2015, India 2014, DR Congo 2012, Ethiopia 2015. Within-country T10/B50 (Chapter 2 data-F2.4): 14.8 (2010), 15.1 (2015, 2017, 2020). Replicates Auditor A.

Why this severity. This is an important undisclosed caveat for any 'current' level and for the recent-trend narrative. The 1980-2020 rise of within-country inequality and the 2021 levels are probably not much affected; a later WID vintage revises 2021 global income concentration up, not down.

The authors' response. Box 1.4 (p.45) says that 2021 growth figures are 'largely projections' and that the estimates 'assume that inequality within countries in 2020 has not been affected by the Covid crisis'. Ch.4 note 2 states that '2021 values are based on estimations'. The age of the underlying country distributions is not given.

Why it matters, in detail. Dividing a CO2-only stock (300/900 GtCO2) by an all-GHG flow (~50 GtCO2e a year, excluding LULUCF) understates the time to depletion. With CO2-only flows (roughly 40 GtCO2 a year including land use, the figure Auditor A cites) it would be about 7.5 and 22.5 years. With the author's own later all-GHG budgets (500/1,300 GtCO2e) it is about 10 and 26 years, and the per-person budgets are 1.9 and 4.9 tCO2e rather than 1.1 and 3.4. Per-person budgets are therefore understated by 31-42%, and budget multiples (e.g. regional footprints relative to the 2C budget) overstated by 44-73%.

Evidence. 300/50.1 = 6.0 years and 900/50.1 = 18 years, as printed; 300/265 = 1.1 t and 900/265 = 3.4 t (Table 6.2). IPCC AR6 WG1 SPM Table SPM.2 defines these as CO2 budgets. Chancel's later Nature Sustainability SI (pp.9-10) states that the 300/900 figures 'only reflect the remaining CO2 emissions budget and do not include remaining non-CO2 gases'. It builds all-GHG budgets of 500 and 1,300 GtCO2e, giving 1.9 and 4.9 tCO2e per person per year, or about 10 and 26 years at 50 Gt a year. Found by both auditors (Codex I2, A I10).

Why this severity. The illustrative budget figures in the carbon chapter are off by roughly a third to three-quarters: years to depletion, per-person budgets and the multiples derived from them. The measured emission shares are unaffected, and so is the qualitative conclusion that top emitters are far above any sustainable level.

The authors' response. Chancel's 2022 Nature Sustainability supplement (pp.9-10) recognises the scope mismatch and gives all-GHG budgets of 1.9 and 4.9 tCO2e per person. The WIR 2022 text, including the March 2023 revision, still contains the original comparison.

Why it matters, in detail. The targets are territorial (NDC basis) and the baselines are footprints. For net importers such as France (footprint/territorial 1.33) the required cuts are overstated: the pledge implies -26%, not -45%. Scaling the target to a footprint basis (about 6.4 t) puts the French bottom 50% (4.95 t) about 22% below target and the middle 40% needing -31% rather than -48%. For net exporters such as China, the permitted increase is overstated (+25% vs +9%). The conclusion that rich countries' bottom-50% footprints are at or below per-capita targets is unchanged.

Evidence. Chapter 6 workbook data-F6.10a-d gives 2019 baselines of FR 8.66, US 21.13, IN 2.17, CN 8.02 t and 2030 targets of FR 4.8, US 10, IN 3.7, CN 10 t. Snapshot territorial per-capita emissions (kntghg999i, 2019) are FR 6.53, US 19.75, IN 2.42, CN 9.16 t. Report p.130 says 'In France, the pledge amounts to a 45% reduction'. Replicates and extends Auditor A.

Why this severity. The required reductions in a summary-level secondary result are misstated by up to about 20 pp (France) and 16 pp (China). The qualitative message that the bottom half in rich countries is at or near the targets holds, or strengthens.

Why it matters, in detail. A broken MER for a hyperinflation economy makes all Venezuelan values at MER about 20 times too high. In MER-based tabulations Venezuela becomes Latin America's richest country by household wealth and supplies close to half of the region's dollar millionaires. Correcting it would cut the region's millionaire count by about 0.6-0.8m, and its MER national income in 2019-2020 by about 40%. Global MER wealth shares and the global wealth-tax revenue are affected by about 1%.

Evidence. Adjudicator check_venezuela_mer.py. Using WID wealth thresholds at MER, Venezuela accounts for 678,000 of 1.52m Latin American adults above $1m (45%) and $1.93tn of $5.14tn. At a MER of 1.2 times its PPP rate it would account for about 5,500 adults. WIR Table 7.3e (data-T7.3e) reports 1,930,728 adults with $5,815bn; the WID regional XL-MER distribution gives 1.62m with the same method. Global MER shares are barely affected (T10 82.28% vs 82.37% with a plausible Venezuelan rate). Venezuela's spurious millionaires are about 1.4% of the global count by this method. Replicates Auditor A (741,000 of 1.54m).

Why this severity. A secondary result, the Latin American wealth-tax table (1.9m millionaires, $5.8tn), is likely overstated by roughly 35-45%, and MER-based Latin American income aggregates for 2019-2020 are distorted. Global headline shares and world revenue change by about 1% or less.

Why it matters, in detail. The global aggregates pool country distributions whose top tails and shapes are largely imputed for about half of adults (index at most 4.5; at most 1 for ~18%). Perturbing those countries' top-10% shares by +/-5 pp, holding means fixed, moves global B50 income across 7.5-9.5% and B50 wealth across 1.1-2.7%. Global T10 moves only within 51.1-53.2% (income) and 74.6-76.2% (wealth); +/-10 pp roughly doubles the B50 range. No standard errors or scenario ranges are reported, so the precision of the bottom-50% headlines is overstated. The direction of the error is not identified: WID argues that survey-based shapes understate top incomes.

Evidence. Adjudicator transparency_sensitivity.py (country aggregation, 2021, PPP; countries with index at most 4.5). - Income, baseline B50 8.37 / T10 52.20: 5 pp lower 9.48 / 51.09; 5 pp higher 7.48 / 53.19; 10 pp lower 10.48 / 50.02. - Wealth, baseline 1.93 / 75.36: 5 pp lower 2.71 / 74.59; 5 pp higher 1.11 / 76.15. This replicates Auditor A's one-sided figures exactly and adds the opposite direction.

Why this severity. The top-10% headlines are robust (+/-1 pp). The bottom-50% headlines (8.5% of income, 2% of wealth) could plausibly differ by about a tenth (income) to two-fifths (wealth) of their level, and the report gives no indication of this.

The authors' response. Chancel & Piketty (2021, JEEA 19(6) pp.3055-3056) acknowledge that 'in a number of regions our corrections to raw survey data often rely on a limited set of countries'. They state that this will not affect general conclusions but 'can certainly affect some of the finer decompositions for the more recent period'. The report publishes each country's transparency index in its country sheets.

Why it matters, in detail. The 1820 and 1910 global levels are dominated by within-region dispersion (the between-country Theil share is 12% in 1820 and about 26% in 1900), and that dispersion is unobserved for most of Asia, Africa and the Middle East. A +/-10% mean-preserving change in it (1820 and 1910 only) moves the 1910 global T10/B50 from 41 to 34-49, against 36-38 in 2020, and the 2020/1820 B50 ratio from 0.55 to 0.47-0.64. Chancel & Piketty's (2021) robustness check, flat versus rising within-country inequality over 1820-1910, addresses the trend, not the level comparisons that the Executive Summary emphasises.

Evidence. Adjudicator longrun_sensitivity.py. Baseline T10/B50 from the 8-region percentile file is 18.5 / 41.0 / 53.2 / 36.3 in 1820 / 1910 / 1980 / 2020 (published 18 / 41 / 53 / 38 from WO_hist). Auditor A obtained 34-52 and 0.47-0.66 (without a floor on negative incomes).

Why this severity. The direction of the long-run story (a rise from 1820 to 1910, then persistently high inequality) survives. But whether 2020 is below or about equal to 1910, and whether the bottom-50% share 'halved' since 1820, both fall within the range produced by modest, unobservable changes in historical dispersion, and the report does not quantify this.

The authors' response. Chancel & Piketty (2021, JEEA section 3.4, pp.3046-3047) report variants of the 1820-1910 evolution of within-country inequality. They write that 'this has relatively little impact on the overall pattern', because the 1820-1910 rise is driven by well-documented between-country divergence. The Executive Summary hedges the 1910 comparison ('seem to be about as great').

Why it matters, in detail. Pooled regional shares mix within- and between-country inequality. For MENA the between-country component is large: the within-only T10 is 51.2% against 57.8% pooled, and T10/B50 falls from 31.3 to 19.6. MENA's top rank therefore reflects the Gulf-vs-rest income gap and oil geography, not the most unequal societies; Latin America ranks first on within-country measures. The heavily imputed Gulf distributions add uncertainty.

Evidence. Adjudicator regional_decomposition.py (2021, PPP, 127 g-percentiles; regional membership validated against WID regional populations): - MENA: country aggregation 57.8% (stored XN 58.1%); within-only 51.2%; T10/B50 31.3 -> 19.6. - Latin America: 55.1% -> 53.9%. - Sub-Saharan Africa: 55.5% -> 49.9%. - Europe (W+E): 35.6% -> 34.5%. - Adult-weighted mean of country top-10% shares: MENA 51.1%, Latin America 53.9%. WIR 2018 Part 2 (Ch.2.10) attributes MENA's inequality mainly to between-country gaps. Replicates Auditor A.

Why this severity. The ranking is correct for the report's stated measure (the pooled regional distribution). The authors documented the between-country driver in WIR 2018, and Europe-lowest and wide regional variation survive. The issue is an omitted, decision-relevant decomposition rather than a numerical error.

The authors' response. Alvaredo, Assouad & Piketty (2019, RIW 65(4)) and WIR 2018 (Ch.2.10) explain that Middle East inequality is driven largely by between-country (Gulf vs other) gaps, and they defend their top-income corrections against survey-based critiques.

Why it matters, in detail. Each regional series is more unequal than the adult-weighted aggregation of its member countries' 127 g-percentiles, while WO matches its country aggregation. Mixing the eight regional series therefore implies a world top-10% income share of 53.1% (a bracket-mean lower bound) instead of 52.2%. The cause cannot be identified from the release (possibly a different vintage or aggregation routine). The C5 regional shares are about 0.2-1 pp too high relative to the country data.

Evidence. Adjudicator headline_and_aggregation.py, regions_vs_countries.py and regions_weight_test.py. Regional membership was reconstructed and validated: member populations sum exactly to each regional npopul992i except Europe. - Income T10, stored vs member aggregation: QE 35.78 vs 35.64 (with Eastern Europe); XB 45.74 vs 44.69; QL 43.41 vs 43.03; XS 54.84 vs 54.36; XL 55.39 vs 55.12; XN 58.12 vs 57.79; XR 46.71 vs 45.84; XF 55.71 vs 55.51. - Wealth T10: QL 69.03 vs 67.67; XR 73.25 vs 71.82; XN 76.61 vs 75.44. - The gaps persist with total-population weights. - Regional mixture: world income T10 53.11%, B50 8.27% (WO 52.17%, 8.40%).

Why this severity. Regional top-10% shares are about 0.2-1 pp higher than their member-country data imply, and regional rankings are unchanged. The world headline is consistent with the country data. This is a reproducibility and consistency defect with small numerical consequences.

Why it matters, in detail. The WO and regional macro totals (npopul992i, mnninc999i) omit 18 Eastern European countries, about 1.9% of adults and 2.6% of PPP national income, while QE's and WO's income distributions appear to include them. Group averages for 'Europe' are therefore about 8-9% too high relative to a Western-plus-Eastern population. World averages differ by under 1%. Wealth-tax revenues expressed as % of world income are about 2% too high if Eastern European millionaires are in the tax base. Whether the wealth WO includes them is ambiguous: the Lorenz fit is slightly better without them.

Evidence. Adjudicator eastern_europe_coverage.py. - 2021 adults: all countries 5,269.1m; excluding the 18 countries 5,169.4m; WO 5,170.3m. - National income at PPP: 88.64, 86.37 and 86.38tn respectively. At MER: 66.75, 65.40 and 65.40tn. - The same pattern holds in 1995-2020. - QE wealth T10: 58.52% stored, 58.39% for Western plus Eastern Europe, 56.02% for Western Europe alone. - Including the 18 countries changes global country-aggregated shares by at most 0.4 pp (income T10 52.20 vs 52.59; wealth T10 75.36 vs 75.58). - WO income Lorenz gap: 0.06 pp with these countries, 0.45 pp without. - The live WID world file (retrieved 2026-10-25) gives 3,389m adults in 1995, against 3,299m in the snapshot and a snapshot all-country sum of 3,391m.

Why this severity. Headline shares move by at most 0.4 pp. Affected are the world averages (EUR 16,700 income, EUR 72,900 wealth; under 1% off), Europe's average income relative to the world (about 2.15 instead of about 1.96 with Eastern Europe), and denominators expressed as '% of global income' (about 2%). No conclusion changes.

Why it matters, in detail. The 50 -> 40 figures are the global interpersonal T10/B50 for 2000-2020; the between-country T10/B50 fell from 15.1 to 8.7 over the same period. The mislabel understates between-country convergence and inflates the apparent between-country gap by more than three times. It does not affect the decomposition results (Theil between-country share 56.9% in 1980 -> 32.0% in 2020).

Evidence. Summary workbook data-F5: global T10/B50 50.42 (2000), 38.44 (2020). Chapter 2 workbook data-F2.4: between-country 21.8 (1980), 15.1 (2000), 8.7 (2020); within-country 8.6 -> 15.1 (1980-2020). The wording persists in the 2022 and 2023 editions. Found by both auditors (Codex I3; A I14d).

Why this severity. This is a labelling error in the summary. The underlying Chapter 2 analysis and the stated conclusion (within-country now exceeds between-country inequality) are correct, and the true between-country decline is larger than the mislabelled one.

Why it matters, in detail. Each item is a mislabel or transcription error relative to the released data: a factor-10 threshold typo, a fractile label wrong by a factor of 10, a 0.1 pp rounding inconsistency, a wrong denominator in a caption, and denominator labels on the regional tables. Underlying series and shares are unaffected.

Evidence. Snapshot (WO sptinc992j p0p50 2021 = 0.0840; thweal992j p99.99p100 1995 = 6,922,536); Summary workbook data-F6 (0.1203), data-F10 (top-0.01% series labelled 'top0.1', 7.29% -> 11.11%; billionaire share 3.34%); Chapter 3 workbook data-F3.1 (world public wealth 0.874 and 0.753 of NI); printed Table 7.3a headers. Found by A (I14) and Codex (I6); checked by the adjudicator.

Why this severity. Presentation and transcription errors. The correct values are in the data and nearby text, and no conclusion changes.

The authors' response. Some figure notes were corrected in the March 2022 Executive Summary. Items (b), (c) and (f) persist in the March 2023 full report.

Why it matters, in detail. These are text-data mismatches in thresholds, growth rates, fractile labels, regional population shares, glossary definitions and timing statements. None of them alters the series from which the core findings are computed.

Evidence. Report text (Dec 2021, with printed pages identified); snapshot values via wealth_growth.py and macro reads; Chapter 6 workbook T6.5, T6.6 and data-T6.6. Most items were found by Auditor A (I15), who also lists others (55,200 vs 51,700 adults; Fig.1.14b legend; copied figure notes in Ch.8-9) that the adjudicator did not re-check.

Why this severity. Each item is individually small and the underlying data are correct. Collectively they show limited checking of the text against the data.

The authors' response. Some wording was revised in the March 2023 edition (e.g. North America 'one of the most unequal'); several items persist.

Why it matters, in detail. Against the report's own 3.4 t 2C budget, bottom-50% footprints exceed it by 35% (Russia & Central Asia), 50% (Europe) and 185% (North America). The statement would be closer to correct with an all-GHG budget (4.9 t), which the report does not use.

Evidence. Executive Summary Fig.15 data; Table 6.2 (3.4 t); report p.123 text. With the author's later all-GHG budget (4.9 t, see F8), the Russia statement would hold and the European one would nearly hold. Found by Auditor A (I11); checked by the adjudicator.

Why this severity. An internal contradiction in one descriptive sentence. The distributional carbon estimates are unaffected.

Why it matters, in detail. Compounding the arithmetic mean of volatile annual growth rates overstates deterministic continuation of historical growth: the top 0.1% grows 4.15% a year against a 4.02% CAGR. Over 79 years this raises the top-0.1% share in 2100 by about 1.8 pp (30.9% -> 32.7%) and makes the crossover fall inside the century. The divisor slip changes shares by about 0.1 pp.

Evidence. Adjudicator projection_2100.py, a translation of Chapter4.do F4.6 run on the snapshot. - Published implementation: top 0.1% 32.72% and middle 40% 30.47% in 2100; first crossover in about 2075 (matches workbook data-F4.6). - Divisor fixed: 32.61% / 30.37%. - Compound annual growth (top 0.1% 4.02% vs arithmetic 4.15%; middle 40% 3.84% vs 3.87%) with the divisor fixed: 30.91% / 31.26%; crossover in about 2107. Replicates Codex I4.

Why this severity. The projection is explicitly hypothetical, and under the more natural compound-growth reading the two groups are nearly level in 2100, with the crossover about seven years later. The qualitative message survives.

The authors' response. The report labels the projection a 'hypothetical scenario' (p.95).

Why it matters, in detail. Linear-regression extrapolations produce shares below zero for one country, and over two-thirds of country-years are modelled. The GDP-weighted global aggregate (about 35%) is dominated by countries with LIS or EU-SILC data and is robust to clamping the negatives.

Evidence. wir_fe_labinc_share_210923_clean.xlsx (sheet WID_countryestimates: 23 negative values; Fig5.3._Global: 0.3064 -> 0.3471; population-weighted 0.2747 -> 0.2855). Codex: setting the negatives to zero changes a population-weighted global series by at most 0.0075 pp. Found by A (I16); treated by Codex as a diagnostic.

Why this severity. The imputation share is disclosed and the negatives have a negligible effect on the global aggregate. Country-level and population-weighted figures are more fragile.

The authors' response. Neef & Robilliard (2021, WIL WP 2021/22) and Box 5.1 disclose the modelling and urge caution at country level.

Why it matters, in detail. Mixed vintages across released workbooks (Dec 2021 code, Mar 2022 workbooks) leave two competing versions of Table 1 and of the scenario-1 schedule. Together with the missing tax input file, the wealth-tax results cannot be traced to a single computation.

Evidence. Summary-2022-03.xlsx sheet T1; WIR2022TablesFigures-Chapter7.xlsx sheets T7.2 and data-T7.1; report p.140. Found by A (I17); T7.2 discrepancy added by the adjudicator.

Why this severity. The numerical differences are small (about 4% in billionaire counts), but the released files do not uniquely define the published tables.

Why it matters, in detail. The European ICU density is overstated by about 1.5 orders of magnitude (4,000 vs roughly 115-300 per million), which inflates the stated Africa-Europe gap by a similar factor. The error is inherited from a secondary source.

Evidence. Report p.174; Chowdhury & Jomo (2020) full-text XML ('about 4000 per million in Europe'); Rhodes et al. (2012) abstract (11.5 per 100,000 in 2010-11). The internal comparison with the US figure was added by the adjudicator. Found by Codex (I7).

Why this severity. A peripheral illustration in a policy box. The general point that health capacity is very unequal is unaffected.

Why it matters, in detail. The US top-1% wealth share is 2-3 pp lower under heterogeneous-return capitalisation (Smith-Zidar-Zwick), and CBO-based progressivity measures contradict the flat effective tax rates in Fig.8.1. Both critiques are partly accepted and partly disputed in the literature. Their effect on world shares is a fraction of a percentage point.

Evidence. Summary workbook data-F11 (US top 1% 35.4% in 2020; 42.5% in 1910); corpus entries us-wealth-taxes-03 and -06 (heterogeneous returns; Smith-Zidar-Zwick 2023 QJE), us-wealth-taxes-12 and -13 (Splinter; Kopczuk on incidence), us-income-08 and -14. Raised by Auditor A (I18).

Why this severity. The qualitative US claims are weakened but not reversed. The US is about 5% of the world's adults, so the global headlines are hardly affected.

The authors' response. Saez & Zucman revised their US series in response (e.g. heterogeneous fixed-income returns in later vintages) and argue that remaining gaps concentrate in the top 0.01%. See the corpus entries.

Why it matters, in detail. One proportional haircut, independent of the marginal rate, is inconsistent with semi-elasticities of taxable wealth that make erosion rise with the tax rate (about 8-13 in US scorings; larger in Swiss evidence). Revenues for the steep scenarios, especially scenario 3's 5.3% with 50-90% top rates, are therefore upper bounds. The effect on scenario 1 is modest; the report's own 40%/10% variant gives 1.2%.

Evidence. Report pp.140-142. Corpus entries us-wealth-taxes-20 / optimal-tax-behavior-17 (Penn Wharton Budget Model 2019: semi-elasticity -13 vs -8, erosion that should scale with the marginal rate), optimal-tax-behavior-19 (Brulhart et al. 2022), optimal-tax-behavior-20 (Advani & Tarrant 2021), and us-wealth-taxes-18 / optimal-tax-behavior-16 (Kopczuk 2019). Raised by Auditor A (I13).

Why this severity. For the headline scenario 1 (rates 1-3.5%) the uniform haircut is within the range of the literature, and the report shows a high-evasion variant (1.2%). The concern bites mainly for scenarios 2-3, which are not the Executive Summary headline.

The authors' response. The report (p.142) states that behavioural responses 'obey no law of nature – they are governed by tax policy choices', provides 0%/0% and 40%/10% variants, and states that scenario 3 cannot raise such revenue for ever (p.140).

Why it matters, in detail. With the top tail pinned to Forbes, a proportional Forbes overstatement of x% would raise the billionaire share by about x% (3.5% -> roughly 1.8-2.5% if Forbes were 1.4-2 times too high), with smaller effects on the top 0.01% and top 0.001%. Their growth since 1995 would be roughly preserved if the bias were constant. The direction of the bias is disputed and unresolved (us-wealth-taxes-11).

Evidence. WIL TN 2021/16 section 2.2 ('we rescale the top of the distribution with Forbes data'); Summary workbook data-F10 billionaire series (1.08% in 1995, 2.38% in 2019, 2.20% in 2020, 3.34% in 2021); corpus entry us-wealth-taxes-11 (verified; Raub et al. read second-hand there). Adjudicator coverage check: neither auditor examined this.

Why this severity. This bears on the level of the billionaire and top-0.001% shares, not on their rise if any Forbes bias is roughly stable. The top-10% and top-1% headlines are hardly affected.

The authors' response. Saez & Zucman (2019 reply to Summers-Sarin; 2020 JEP p.9; 2022 NBER w30396) read the estate-Forbes gap as estate-tax avoidance and use SEC filings as a check on billionaire equity (as summarised in the corpus). The WIR does not discuss Forbes uncertainty.

Why it matters, in detail. Comparing period means cannot separate tax progressivity from post-war reconstruction and catch-up, demography and productivity trends. Better-identified time-series evidence finds positive short-run output effects of top-rate cuts. 'Boosted' is a causal claim that the evidence shown does not establish, although the narrower claim that growth did not accelerate after the cuts is descriptively accurate.

Evidence. Report pp.168-170. Related corpus entries: optimal-tax-behavior-10 (Reynolds 2019 on growth comparisons) and optimal-tax-behavior-11 (Mertens & Montiel Olea 2018: top-rate cuts raise short-run output). Raised by Auditor A (I19).

Why this severity. An interpretive policy claim outside the report's quantitative core; it does not affect any measured finding.