← All papers
Emmanuel Saez, Gabriel Zucman · Preliminary draft, UC Berkeley, November 2019 (listed in CV)
A National Income Tax
15 problems found: 5 moderate · 10 minor
What the paper claims
- Design and base. A single-rate tax with no deductions on all national income: employers remit it on full labour cost (all fringe benefits, no cap); businesses pay on profits (economic depreciation, no deduction of taxes); interest and foreign income received by individuals and nonprofits are taxed; US dividends, retirement income and transfers are not; owner-occupied imputed rent is excluded and mortgage interest is not deductible. The base is 'nearing 100%' of national income before evasion and 'slightly over 90%' after existing misreporting of about 7% of national income.
- Revenue. Because the base is broad, low rates raise large revenue: a 6% NIT raises 5.6% of national income; a 10% rate raises about 8% after offsetting effects on other tax bases; combined with greater taxation of the rich the package raises about ten points of national income.
- Distribution and comparison with a VAT. The NIT 'increases tax rates almost uniformly across income groups' and, because it exempts transfers and does not exempt saving, is 'much more progressive than a VAT'. The VAT base is only ~40% of national income (France and Germany raise ~8% of national income at 20%/19% standard rates), so the US would need a 15% VAT to raise 6% of national income.
- Health-care application (companion materials; may or may not be in the unreleased draft). A 4.5% NIT would fund standard insurance for all workers now covered through employers; 6% would also cover the 30 million uninsured ('true universal health insurance'); replacing employer premiums (a 'privatized poll tax' of 7% of national income) with the NIT makes most Americans better off: in the book's reform all groups up to ~P95 pay less and the median rate falls from 38% to 28%; worked examples show take-home pay rising 20-22%; 'with a 6% national income tax, 90% of the population would be better off'.
Every problem found
1moderateconfirmed · newly identified here · overclaiming
The authors say a 6% tax on all income would by itself pay for health insurance for everyone. Their own spreadsheets show it raises roughly two-thirds to four-fifths of the cost they assign to it. Their overall plan still balances, but only because other new taxes chip in, or because of extra income tax on higher wages that the authors never count.
Technical detail and evidence
Why it matters, in detail. Gross yield per NIT point is 0.907% of the simulator denominator (NI plus 'pure' capital gains). It is 0.794% net of the workbook's PIT/CIT offset rule (calculator V and W scaled by 1-R), and about 0.72% if payroll taxes are also offset. At 6% that gives 5.44% gross, 4.76% net and 4.34% net of payroll. Against Table 9.2's 6.0% health line this is 91%/79%/72%; against premium replacement (6.92%) it is 79%/69%/63%. The shortfall (≈1.3-2.6 pts of NI net) is the same order as the income and payroll tax on converted premiums (2.0-2.3% of the denominator; CBO/JCT federal exclusion ≈1.55% of 2019 NI). So the claim holds only under an accounting that the authors never state and that contradicts their distributional figures.
Evidence. TOI e-book p.167 (Table 9.2) and p.168; SZ2019MainFigures.xlsx F9.2!C12 (5.6%), C16 (6.0%), A24 (offset note); Tax Simulator Description printed p.11; SZ2019Slides.pdf p.35; SZ2020HealthCare.pdf p.5. Adjudicator scripts: base_and_revenue.py (6%: 5.44/4.76/4.34% of the simulator denominator), converted_premium_taxes.py (rate needed 7.63% gross, 8.72% net), simulator_reimpl.py (book plan surplus 0.38% at 6%, 1.14% at 7%).
Why this severity. A headline policy claim of the companion materials is overstated: the 6% NIT's yield falls 7% (gross) to 22-28% (net) short of the book's own 6.0% health cost, and 21-37% short of replacing current premiums. The plan-level total in Table 9.2 still balances, and a consistent accounting that taxes converted premiums (F2) would roughly close the gap, so the qualitative case for an NIT-funded system survives.
2moderateconfirmed · previously known · methodology
Today, health premiums paid by employers are not taxed. If they are turned into extra pay, as the plan requires, that pay would be taxed like any other wage. Workers would keep roughly a quarter to two-fifths less than the authors' examples suggest. Most middle-class workers would still come out ahead, just by less.
Technical detail and evidence
Why it matters, in detail. Holding employer labour cost fixed, a premium P becomes cash P/1.0765 and bears employee FICA, federal (12-22%) and state (0-4.5%) tax: extra tax ≈25-30% of P for middle earners. Book plan (NIT 7%), P50-60: reform rate 28.9% -> 32.1-32.7% (cut 9.5 -> 5.7-6.3 pts); P90-95 turns from gainer to loser. The extra revenue is 2.0-2.3% of the denominator. A revenue-neutral NIT is 4.0-4.4%, which puts P50-60 at 29.9-30.1%. Worked examples: slides +$8,880 -> +$5,360-6,490; op-ed +$11,100 -> +$6,700-8,110; book opt-out +$8,000 -> +$4,930-5,970 counting taxes on the new wages only, or +$2,900-3,500 also netting the NIT.
Evidence. taxsimulator.xlsx calculator!T8:T22, W8:W22, K8:K22 (formulas read as text); TOI e-book p.168; SZ2020HealthCare.pdf p.6; Guardian op-ed 25 Oct 2019 (PNHP repost: secretary '$50,000 ... would take home $61,000'). CBO, Federal Subsidies for Health Insurance Coverage for People Under Age 65: 2019 to 2029 (May 2019), Table 2: tax exclusion for employment-based coverage $283bn in FY2019, income plus payroll. Adjudicator: converted_premium_taxes.py, bookplan_taxed_conversion.py, worked_examples_adj.py.
Why this severity. Every published take-home figure is overstated by roughly 25-40% (about 60% for the book's opt-out example). With the NIT held at the plan rate, the median rate cut in Fig. 9.3 shrinks by about a third. But the omission is revenue-symmetric: in a revenue-neutral variant the median gain shrinks by only ~11-13% and 95% of charted adults still gain. So magnitudes change but the conclusion does not.
The authors' response. None located. The simulator note (fn.11) says only that in the short run the wage pass-through 'would have to be legislated'.
3moderateconfirmed · newly identified here · internal inconsistency
The book says its plan uses a 6% tax. The chart showing who wins and loses, and the companion website the book points to, actually use 7%. If anything the chart is less flattering to the plan than a 6% version would be, but readers are not told that the two differ.
Technical detail and evidence
Why it matters, in detail. NIT revenue is 6.35% vs 5.44% of the simulator denominator. The book plan's simulator surplus is 1.14% (7%) or 0.38% (6%) against ≈2.8% needed for the uninsured at $8,000 (≈1.3%) plus education (1.5%). The plotted P0-P95 reform rates are 0.68-0.88 pts above a 6% plan. The text's median sentence (38% -> 28%) cannot tell the two rates apart (see rejected claims).
Evidence. SZ2019MainFigures.xlsx DataF9!I4:I18; taxsimulator.xlsx calculator!EC8:EC22 and interface!G10; simulator_reimpl.py (validation: current law exact; BOOK PLAN max |diff| 6e-17); taxjusticenow.org main-es2015 bundles (Wayback 20191018114309; current) preset objects; TOI e-book pp.167-169.
Why this severity. It does not weaken the distributional conclusion, since a 6% version would show lower middle-class rates. But it is an undisclosed discrepancy in the plan's headline rate worth ≈0.9% of NI (≈$160bn a year). It also shows that the website version, which pays for premiums at their current cost, needs 7% and still cannot fund the book's other spending.
4moderateconfirmed · newly identified here · documentation gap
The paper the book cites for 'complete details' has never been made public, and the code that computes who would pay the tax has not been released. Outsiders can check the totals roughly but cannot check how the authors built them.
Technical detail and evidence
Why it matters, in detail. The D4 government-interest factor (0.5) alone moves the base by ±1.7 pts of NI (87.2-90.6%). The 3.4 pt gap between D4 (88.9%, 2017) and the simulator (92.3% of pre-tax income, 2018) cannot be decomposed. The per-group base ratios (0.69 at P0-10 to 0.98 at P70-90) cannot be traced to components such as pension-fund income, nonprofit surplus, foreign income or the allocation of misreporting.
Evidence. Zucman CV p.8; TOI p.182 n.15 and p.188 bibliography; taxsimulator.xlsx data!A19; SZ2019.zip program list (16 files, none with that name or mentioning nitaxbase; codex source_checks.json); SZ2019AppendixTables.xlsx TD4!A31:A37, D19:D21; Wayback CDX listings (claude); base_and_revenue.py.
Why this severity. The aggregate base can be approximately rebuilt (D4 reproduces exactly), so the headline is not in doubt. But the composition and distribution of the base, which drive C2-C3, cannot be verified, and the cited 'complete details' do not exist publicly seven years on.
5moderatelikely · newly identified here · speculative assumption
The plan only adds up if public insurance costs about $8,000 per insured worker, against about $13,000 for today's employer plans. The authors give no source for that number. If coverage cost what it costs today, the tax would need to be closer to 8-9% than 6%.
Technical detail and evidence
Why it matters, in detail. Each $1,000 per covered worker ≈ $87bn ≈ 0.48% of 2019 NI ≈ 0.5-0.6 NIT points. Costing 87.3m covered workers at $13,000 and 30m uninsured at $8,000 gives ≈7.55% of 2019 NI, which needs ≈8.1% gross or ≈9.3% net, against 6% claimed. At the stated parameters, $8,000 x 117.3m ≈ 5.15% of 2019 NI, below Table 9.2's 6.0%, so the table also contains unspecified items. CBO's single-payer modelling finds lower provider rates and administrative savings largely offset by higher utilisation at the national level. A 38% per-worker cost cut for the employer-covered population is therefore at the optimistic end, and it is not stress-tested.
Evidence. SZ2019MainFigures.xlsx F9.2!B16:C17; TOI e-book pp.167-168; ch.5 n.6 (p.178); ch.9 notes 1-16 (p.182) cite no source for $8,000. simulator_reimpl.py (book-plan surplus 1.14% of the denominator at 7%, 0.38% at 6%). Context: CBO, 'How CBO Analyzes Proposals for a Single-Payer Health Care System' (P. Swagel, 10 Dec 2020; Wayback copy): the illustrative single-payer options change 2030 national health expenditures by between -$0.7tn and +$0.3tn.
Why this severity. The NIT base and revenue (C1-C3) do not depend on this assumption, but the '6% is enough' health claims do. At the current per-worker premium the NIT needed would be about 8-9% rather than 6%, so the magnitude of C4 is materially uncertain.
6minorconfirmed · newly identified here · methodology
A new tax on wages shrinks the base of the existing payroll tax, so some of the new money is lost elsewhere. The authors count this for income taxes but not payroll taxes, which overstates net revenue by about a tenth.
Technical detail and evidence
Why it matters, in detail. Net yield at 6%: 4.34% vs 4.76% of the simulator denominator (0.49 pts lower at 7%). '10% raises 8%' becomes ≈7.2%. The size depends on incidence: if the labour component fell on profits, the offset would appear in corporate and income taxes instead.
Evidence. taxsimulator.xlsx calculator!V8:W22 vs K8:K22; SZ2019MainFigures.xlsx F9.2!A24; CBO budget option 54820 (Dec 2018, Wayback 20190605); base_and_revenue.py.
Why this severity. Lowers net NIT revenue by about 9% (0.42 pts of NI at 6%). Below the threshold that would change the qualitative revenue claim, but it adds to the F1 shortfall.
7minorconfirmed · newly identified here · internal inconsistency
The authors' own estimates of how much income the tax would cover range from about 89% to 94% of national income, depending on the document. The differences are small and do not change the main point, but the documents do not agree with each other.
Technical detail and evidence
Why it matters, in detail. At 6%, gross yield is 5.33% of NI on the D4 base and 5.54-5.59% on the simulator base. The simulator base, which drives all published revenue and distribution figures, is at the high end of the authors' own range. The source of the 3-4 pt gap cannot be identified (F5).
Evidence. SZ2019AppendixTables.xlsx TD4!E23; Tax Simulator Description printed p.11; SZ2019Slides.pdf p.35; Table 9.2; TOI appendix (SZ2019Appendix.pdf) pp.10-11 on the pure-capital-gains denominator; base_and_revenue.py (16,334.1 / 18,012.9 / 17,696.8 / 17,546); simulator_reimpl.py; claude dina_gperc_nitbase.py output.
Why this severity. All estimates are within ±3 pts of 91% of NI, so 'slightly over 90%' and the broad-base conclusion stand. Gross revenue at 6% ranges only from ≈5.3% to 5.6% of NI.
8minorconfirmed · newly identified here · overclaiming
A slide claims 90% of people would be better off, but the chart on that same slide shows the bottom tenth and the top 5% paying more. The scenario shown also does not raise enough money to replace what people pay now. Done consistently, about 70-85% of adults gain.
Technical detail and evidence
Why it matters, in detail. The chart line equals current rates minus premiums plus the gross NIT (6 x base/income). With PIT/CIT offsets the scenario's surplus is -2.16% of the denominator. Revenue-neutral NIT replacing premiums alone: 8.72%.
Evidence. SZ2020HealthCare.pdf ('Financing Medicare for All', 25 Jan 2020) p.7 (chart inspected); slide2020_scenario.py (6%: net surplus -2.16% of the denominator; groups better off P10-P95 = 85%; 8.72%: P20-P90 = 70%).
Why this severity. A later presentation document, not the draft. A large majority of the middle class gains in every variant, but the headline overstates the share by about 5-20 points.
9minorconfirmed · newly identified here · overclaiming
The tax takes a slightly smaller share of income from the very poorest and the very richest than from the middle, so it is not quite as even across income groups as described. The difference is modest.
Technical detail and evidence
Why it matters, in detail. Base/income ranges from 0.69 to 0.98 across groups. The top-end dip comes partly from the PSZ-style allocation of NIPA misreporting to business income at the top (an audit-based allocation as in Auten-Splinter would make the NIT more uniform) and partly from the pure-capital-gains denominator.
Evidence. Tax Simulator Description printed p.11; taxsimulator.xlsx data!AT/O/E (base_and_revenue.py by-group ratios: base/denominator 0.69 at P0-10 to 0.98 at P70-90, 0.78-0.81 for the top 0.1%; base/pre-tax income 0.87-0.89 for the top 1%); Codex R5 (gross increments 6.9-9.8 pts at 10%).
Why this severity. This is a descriptive overstatement. The NIT is still close to proportional over most of the distribution and far less regressive than a consumption tax, so the conclusions are unaffected.
10minorconfirmed · newly identified here · arithmetic or transcription
A small rounding slip in a footnote; it changes nothing.
Technical detail and evidence
Why it matters, in detail. 7.13% printed as 7.2%. The Oct-2019 vintage gives 7.15%; the current vintage gives 6.78% after a downward revision of proprietors' misreporting, which would slightly raise the base.
Evidence. TOI e-book p.182 n.15; SZ2019AppendixTables.xlsx TD4!E14; ALFRED vintage 2018-12-01 (BA07RC1A027NBEA 86.153, B2636C1A027NBEA 672.300, B1633C1A027NBEA 367.250, A032RC1A027NBEA 15,782.997 for 2015).
Why this severity. The difference is 0.07 pts of NI; the components are exact.
11minorconfirmed · newly identified here · arithmetic or transcription
One sentence in the book says the opposite of what the authors mean: workers gain if their premiums are higher, not lower, than 6% of their pay.
Technical detail and evidence
Why it matters, in detail. The sign of the inequality is reversed; the substantive claim (most covered workers pay premiums above 6% of labour cost) is plausible given $13,000 average premiums but unverified.
Evidence. TOI e-book p.168.
Why this severity. This is a wording error; the surrounding worked example is computed with the correct logic.
12minorlikely · newly identified here · internal inconsistency
The authors' model counts about 7% of national income as health premiums paid through employers, but their own book puts those premiums at about 6.2%. The extra seems to come from other kinds of private insurance. This makes today's burden on middle-class workers, and so their gain from the reform, look somewhat bigger than it is.
Technical detail and evidence
Why it matters, in detail. Premiums removed: 6.92% vs ≈6.1% of the denominator for ESI only. Fig. 9.3's 'about thirteen points' of premiums at the median would be ≈11.4. Replacing all private insurance, as single payer would, is arguably the right funding target, but the distribution assigns non-employer premiums and government subsidies to employer-covered workers.
Evidence. taxsimulator.xlsx calculator!AS8:AU22 (AT/AS = 2.3112 in every row; AT24 = $5,752 per adult x 216.69m); CMS NHE Dec-2018 release Tables 05-6 and 24 (cached by claude); TOI ch.5 nn.6, 9 (p.178); Tax Simulator Description p.11; SZ2020HealthCare.pdf p.3. Projected to 2018 at NI growth, ESI ≈ $1,093bn ≈ 6.2% of NI.
Why this severity. This inflates the current-law premium burden of covered workers by ≈14%, about 1.6 of the 13.0 points at P50-60. A revenue-consistent rescaling would lower the NIT needed by ≈1 point, so the net effect on middle-class gains is under 1 point. The identification is inferred from a numerical match because the building code is unpublished.
13minorplausible · previously known · methodology
The revenue numbers assume nobody changes their behaviour, and that profits of US firms abroad can be taxed. Both are optimistic, but for a low, flat tax the likely loss is a modest slice of the revenue.
Technical detail and evidence
Why it matters, in detail. Illustratively, an elasticity of 0.1-0.25 on Δln(1-τ) ≈ -0.10 reduces taxable income by 1-2.6%, costing ≈0.4-0.9% of NI across all taxes. Profit shifting would erode the ≈12.5%-of-NI corporate-profit component unless minimum-tax rules apply.
Evidence. Tax Simulator Description s.3.1; TOI e-book p.166; SZ2019AppendixTables.xlsx TD4!D21; BEA NIPA Table 1.7.5 (Sept 2026 release): income receipts from the rest of the world $1,031bn in 2017. An illustrative elasticity of taxable income of 0.1-0.25 implies ≈0.4-0.9% of NI lost across all taxes (auditor A and adjudicator rough calculations, not sourced estimates).
Why this severity. The assumption is disclosed and standard for a simulator, and responses to a low-rate, broad, withheld tax are probably modest. The plausible loss of ≈7-17% of gross NIT revenue does not change the qualitative claims.
The authors' response. The authors disclose static scoring, argue that avoidance is a policy choice, and condition the profit component on the multinational-tax reforms of TOI ch.6.
14minorplausible · previously known · methodology
Who really ends up paying a tax on company profits is debated. Some of it may come out of workers' pay. If so, the tax would be a little less progressive than shown, but only slightly.
Technical detail and evidence
Why it matters, in detail. 0.25 x 6% x ≈11% of NI (corporate share of the base) ≈ 0.17% of NI. That is roughly +0.2 pts for middle groups and -0.4 pts for the top 1%.
Evidence. Tax Simulator Description pp.3-4; Saez & Zucman, 'Clarifying Distributional Tax Incidence: Who Pays Current Taxes vs. Tax Reform Analysis' (Oct 2019, abstract and s.1); corpus us-wealth-taxes-13 (Kopczuk BPEA 2019 pp.523-524).
Why this severity. Shifting a quarter of the corporate component to labour moves only ≈0.2% of NI of NIT burden, so the near-flat profile and middle-class conclusions are unchanged.
The authors' response. Saez & Zucman (2019, 'Clarifying Distributional Tax Incidence') defend assigning current taxes by statutory/factor incidence and reserve behavioural incidence for reform analysis.
15minorplausible · previously known · overclaiming
Sales-type taxes look harsh on the poor when measured against one year's income, but much less so over a lifetime, and pensions rise with prices. The new tax is still fairer than a sales tax, but probably by less than the book implies.
Technical detail and evidence
Why it matters, in detail. Annual-income incidence overstates VAT regressivity because consumption/income falls with transitory income. CPI indexation shields Social Security recipients, and price-level effects on indexed transfers are not modelled. The book's 'decades or centuries' bequest argument answers only part of the lifetime critique.
Evidence. TOI e-book pp.163-165; Tax Simulator Description p.11 ('the assignment of consumption taxes such as VAT cannot be made with very good accuracy'); Caspersen & Metcalf, 'Is a Value Added Tax Regressive? Annual Versus Lifetime Incidence Measures', National Tax Journal 47(4): 731-746 (1994), doi:10.1086/NTJ41789105 (NBER WP 4387 abstract: 'a VAT in the United States would be proportional to slightly progressive over the lifetime'); corpus us-wealth-taxes-19 (Mankiw BPEA 2019: VAT plus transfers).
Why this severity. Qualitatively the comparison survives, because the NIT also taxes the normal return to saving, which a VAT exempts. But the size of the progressivity gap depends on the incidence framework and is not quantified.
The authors' response. TOI ch.9 (p.163): the rich can postpone consumption taxes 'decades' or 'centuries' by saving and bequeathing, so annual comparisons capture a real injustice.
Claims examined and rejected
Problems one auditor raised that did not survive checking.
- The book's median statement ('from 38% to 28% ... replaced by six points of national income tax') reproduces only with a 7% NIT, corroborating that Fig. 9.3 uses 7%. — The median sentence does not tell 6% from 7%. At the P50 percentile, 7% gives 37.5% -> 28.0%. For the figure's own P50-60 group, 6% gives 38.4% -> 28.0% with 5.6 NIT points ('six points'), while 7% gives 28.9% with 6.5 points. The 7% finding (F4) stands on the exact reproduction of the figure data, not on this sentence.
- The NIT base is 90.68% of pre-tax national income, and the independent DINA reconstruction (88-89% after evasion) is close to it (listed as a check passed). — The simulator's denominator is pre-tax income plus 'pure' capital gains ($18,013bn vs $17,697bn). On the same pre-tax income total the simulator base is 92.3%, so the reconstruction is 3-4 pts lower and agrees instead with Table D4 (88.9%). The discrepancy is recorded in F7 (minor).
- I7 magnitude: at 6% the NIT is 5.8% of pre-tax income for P50-90 but 4.8% for the top 0.1%. — Partly rejected. About half of the top-end shortfall comes from pure capital gains in the denominator, which is not national income. Relative to pre-tax income the top 0.1% pays ≈5.3%. The issue survives as minor (F10) with the corrected magnitude.
- I2 magnitude: the 6% NIT falls short of the health cost by 1.2-2.4 pts of NI (21-34%). — Partly rejected. Those figures compare the NIT with the cost of replacing current premiums (6.2-6.95% of NI), which is the Jan-2020 slides' framing. Against the book's own Table 9.2 cost (6.0%), the shortfall is 0.4 pts gross and 1.3-1.7 pts net (7-28%). The overclaim survives (F1).
- I4 (7% poll tax = total private health insurance) is of moderate severity. — Severity reduced to minor (F8). Once the NIT needed is rescaled to the smaller premium total, the net effect on middle-class gains is under 1 point. The identification rests on a numerical match. And replacing all private insurance is arguably the relevant funding target for a single-payer reform.
- Apart from the documentation gap, no revenue overstatement was demonstrated, and the approximate-uniformity claim stands as stated. — Partly rejected. There are small but demonstrable overstatements in the companion revenue arithmetic: the payroll offset is omitted (-0.42 pts at 6%, F6); Table 9.2's NIT line is 0.16 pts above the simulator in the units of its other lines; and the simulator base is 3-4 pts above the authors' own Table D4 (F7). The health-funding sufficiency claims are overstated in the authors' own accounting (F1). Codex excluded the health package from scope rather than examining it. The uniformity claim survives only as approximate (F10, minor).
- Possible transcription problem in TD4!D20: interest paid by individuals (excl. mortgage) $251.4bn vs BEA B069RC $293.9bn for 2017 (Codex itself did not classify this as an error). — Not an issue. The mapping is undocumented, but a value below total personal interest payments is consistent with excluding interest paid to government, which the NIT would not tax. Substituting the BEA series would raise, not lower, the base (+0.25 pts).
Audit notes
What reproduced. Reproduced: the simulator base (90.68% of its denominator; 92.3% of pre-tax income); Table D4 (88.87% of 2017 NI) exactly; the n.15 misreporting components; the 62% compensation share; '10% raises 8%' (7.94% net); the Table 9.2 wealth, income+corporate and sales lines; the France/Germany VAT figures; and the Fig. 9.3 reform line and the Jan-2020 slide chart, exactly, from a Python re-implementation of the simulator. The Fig. 9.3 line reproduces only with a 7% NIT. New estimates: income and payroll taxes on converted premiums of 2.0-2.3% of the denominator (CBO/JCT federal anchor 1.55% of 2019 NI), and recomputed worked examples. Not reproduced: the draft (unavailable); the per-percentile construction of nitaxbase (taxsimuweb.do unpublished; independent reconstructions land 3-4 pts lower); Table 9.2's 5.6% NIT line in the units of its other lines (5.44%) and its 6.0% health cost from the stated parameters (5.15%); the '6% is enough' claims; the '90% better off' headline; and the 'over 90% of covered workers' statistic (no microdata).
Limits of this audit. The target paper is unavailable, so all findings concern companion materials by the same authors (TOI ch.9 and appendix, the simulator description, workbooks and website, slides, op-ed). The draft may differ from them or resolve some issues. C4 is attributed to the work only through these materials. The per-percentile NIT base and the health variable come from unpublished code, so their composition was inferred, not verified. Magnitudes for taxes on converted premiums depend on assumed marginal rates (federal from the simulator's MTR column, state 0-4.5%, Social Security cap approximated by group). The plausibility of the $8,000 cost was assessed only against the book's own premium figure and CBO's single-payer summary, without health-economics modelling. Behavioural responses were illustrated, not estimated. E-book page numbers are those of the cached PDF conversion, not print pagination. The Jan-2020 slides postdate the draft.