We Checked Our Own Numbers First. Two of Them Were Wrong.

By Common Ladder · August 18, 2026 · 7 min read

The most useful thing that happened while building an interactive tax tool was that two of our own numbers turned out to be wrong.

Neither was a typo. Both had been sitting in the project's research brief for weeks, formatted like every other figure, attributed to a real institution, and repeated in planning documents. They were the kind of number that gets copied forward because it already looks checked.

The two corrections

The first concerns housing vouchers, which is the closest thing this site has to a signature policy. The brief claimed that fully funding the Housing Choice Voucher program would lift 9.3 million people above the poverty line, attributed to the Urban Institute in 2021.

Urban's estimate, published in August 2023 and modelled on 2022 conditions, is that the poverty rate would fall from 14.7% to 12.8% — 6.4 million people. The extra cost is $118 billion a year, gross, with nothing netted out for reduced spending elsewhere. About a quarter of eligible households currently receive any housing assistance at all.¹ No Urban publication we could reach contains the 9.3 million figure.

The second concerns the 2021 Child Tax Credit expansion, the best natural experiment American anti-poverty policy has produced this century. The brief priced one year of it at "about $105 billion." The Joint Committee on Taxation's score, as reported, is $109.5 billion.⁵

A four-billion-dollar difference does not change anyone's mind about the Child Tax Credit. That is roughly the point. A number that survives because nobody would notice it being off is exactly the number that never gets checked.

Why the order matters

Both figures were on their way onto a published page. Both were caught in a verification pass that ran as a separate piece of work before any of the scenario cards existed — a pass whose only rule was that every figure had to come back with the sentence it was read from, quoted, or be marked unverified and dropped.

Run that pass after the build and the incentives invert. By then the number is already in the chart, the chart is already in the layout, and the correction costs a rebuild. The 9.3 million figure would very likely have shipped, because it is a better number than 6.4 million and there would have been a deadline.

The same pass turned up a third case worth naming, because it cuts the other way. This project had stored the 2022 average federal tax rate for the highest quintile from a secondary reading of a Congressional Budget Office report. Two of the three figures in that reading were right; the third was wrong by 2.7 points. The tool now shows CBO's own numbers: in 2022, the lowest quintile paid 1.4% of income in all federal taxes combined, the highest quintile 25.9%, and the top 1% 31.5%.⁴ Checking a source against a source cannot find that. Only the spreadsheet can.

What the tool does

The Tax Ladder is in three parts.

The history runs 1913 to 2026 and is built around one gap. The top statutory rate — the headline number in the law — has ranged from 7% in 1913 to 94% in 1944, and sat at 91% for the whole of the 1950s. Almost nobody at the top paid it. Plotting the statutory rate against the effective rate the very top actually paid turns "they used to tax the rich at 91%" from a slogan into a measurable distance, and that distance is the entire subject of the second part.

The sandbox gives you the levers: top marginal rate and the income it starts at, capital gains, the corporate rate, the payroll cap, ten loophole toggles, IRS enforcement, and swaps to a flat tax, a national sales tax, a VAT, a wealth tax or a financial transaction tax. Every change produces four things at once — the ten-year revenue effect broken into what it raises on paper, what avoidance takes back, and what is left; a signed bar chart of what happens to each economic class, shown with every result; a GDP range, never a single number; and a plain-language explanation of why you got that answer.

Push the top rate high enough and revenue rolls over and starts falling. Where it turns depends entirely on one assumption dial, and the dial is in the corner where you can move it. A revenue-maximising rate is not a recommendation; it is the point past which a government collects less, which is a different claim from what a government should do.

Twelve scenario cards carry the published scores for real proposals — closing the big loopholes, a 70% top bracket, a wealth tax, a flat tax, the FairTax, funding the IRS, scrapping the payroll cap, restoring the 2021 Child Tax Credit, housing vouchers for everyone who qualifies, a carbon dividend, current law, and the 1986 reform as a template. Tapping a card writes its settings into the sandbox and lets the same engine run. A card does not compute its own answer, so a card and the sandbox cannot drift apart.

Where honest people disagree, both numbers ship

Disputed estimates appear as pairs with names attached, never as an average. The share of income held by the top 1% in 2022 is 23.6% or 15.5% depending on whose allocation choices you accept — an eight-point gap between two credible series using the same definition of income. Both lines are on the chart.

The Child Tax Credit card carries a prediction and an observation that point in opposite directions: one team modelled a permanent child allowance and projected a large drop in employment, another measured the actual 2021 payments and found no detectable effect.² Both can be right, because a payment everyone knows will stop in six months is not the policy the first team modelled. The card says so.

The assumption dial that governs the whole engine is labelled with the real modelling shops each setting imitates, so a reader can find out which set of assumptions they have been agreeing with.

What it will not say

The methodology page publishes the list. No tax cut pays for itself. No single-number GDP effect. No year-by-year macroeconomic path. No distributional claim finer than a quintile, plus the top 1%. No certainty about who ultimately bears the corporate tax. No optimal rate.

Those are constraints on the engine, enforced by tests: a sweep across every combination of rate cut, assumption setting and model lens confirms that no reachable state of the tool displays a rate cut raising money.⁶

Why a homelessness site built a tax tool

Common Ladder's argument is that homelessness is a systems problem. The tax code is the system one level up: it decides what the safety net can afford before anyone argues about how to spend it.

The vouchers card is where the two halves meet. Housing Choice Vouchers reach about one in four eligible households; qualifying does not get you a voucher, it gets you a place on a waiting list.¹ The randomised evidence on what a long-term subsidy does for a family in shelter is among the strongest in the field.³ The gap between that evidence and what gets funded is a budget decision, and the sandbox is where you can see what closing it would take.

The tool will not tell you which lever to pull. It will tell you what each one costs, who pays for it, who published the estimate, and what the people who disagree with that estimate say.

Open The Tax Ladder → · Read the methodology and sources

Figures are as of the fiscal years shown on each one. This page and the tool are reviewed when major tax legislation passes, not on a schedule. Found a figure you can't reproduce from its source? Email corrections.commonladder@gmail.com