Worst offenders

The single largest entry on the site by cost, $254B a year, and by the site’s own arithmetic (cost × concentration score) the largest dollar amount flowing to a narrow class of any entry: $223B of it. Realised long-term gains are overwhelmingly reported by households holding large portfolios of appreciated assets outside retirement accounts; the rate preference has no effect on wage income at all.

JCT has never distributed this provision by income class, so there is no published benefit-share figure to cite: it is the largest entry on the site with no official distribution at all. The closest independent evidence is IRS’s own return data: 89.0% of reported net capital gain amounts sit on returns with adjusted gross income over $200,000. That is a share of the raw amount, not of the benefit, and it almost certainly understates the true concentration, since the benefit from a preferential rate is the rate differential itself, and higher earners face a larger gap between the ordinary and preferential rate. This site does not model that gap, so it is not published as a number, but the direction only ever points one way, toward more concentrated, never less.

The most top-concentrated entry on the site by a figure JCT actually published, not one inferred from raw amounts. Table 3 of JCX-45-25 puts 92% of the individual deduction on returns over $200,000, and 61% on returns over $1,000,000 alone. Returns under $100,000 account for about 1%. The deduction’s value rises with the donor’s own tax rate, which is the mechanical reason the two move together: the same gift is worth more, in forgone tax, to the donor already paying the highest rate.

No entry on the site scores higher on structural narrowness than this one. Unrealised capital gains on an inherited asset are simply erased at death rather than taxed, a benefit that requires holding a substantially appreciated asset until death to receive at all, and whose size is proportional to accrued gain, which is itself concentrated in a small share of estates. $66.3B a year, entirely contingent on dying with the asset rather than selling it.

The complement to the other three: those are individual-side provisions with at least a nominal path for an ordinary filer to benefit. This one has none. GILTI and FDII preferences are available only to entities with foreign subsidiaries and cross-border intangible property, a small number of large multinational corporations, full stop. $62.2B a year, and no version of this provision touches an individual return at all.

Different from the other four in kind, not degree: this is not about who the money reaches, it is about whether it does what it was enacted to do. These two provisions exist because Congress judged that letting firms deduct or credit research costs immediately causes more research to happen than would otherwise. That is testable, and it was tested directly: firm-level R&D spending from SEC filings, checked across the exact tax year (2022) research costs stopped being immediately deductible.

No industry-consistent slowdown appears in that year. Three independent series, BEA’s national R&D-to-GDP share, NSF’s survey total, and the SEC firm panel once a single company’s outlier year is set aside, all show growth continuing through the boundary rather than breaking at it. That does not mean the provision is proven not to work: firms had a live, House-passed expectation of a retroactive fix that the Senate blocked by only four votes through most of the period studied, which would dampen a real response without the provision being neutral. The honest reading is that $75.8B a year is being spent on a stated purpose this project could not find evidence for, not that the purpose is disproven. Full method, confounders and the falsifier stated in advance: docs/rd-hypothesis.md .

The method that finds the five entries above also finds this, and reporting it is the same discipline as reporting them: not every large transfer is captured. JCT Table 3 puts 84% of the credit on returns under $50,000 and 92% under $60,000; returns over $100,000 account for less than 0.2%. It is the most tightly bottom-targeted entry on the entire site, $66.9B a year going almost entirely to lower-income working households, which is exactly what the credit was built to do.