The One Number They Have to Get Right

• Key findings
• It is a wall, not a gradient. Earned income: 4,128 rows, 100% exact, 0 bracketed. Liabilities: 7,612 rows, 100% bracketed, 0 exact. Assets: 179,023 bracketed. No schedule is mixed.
• 2.2% of the valued disclosure is a number. The other 97.8% is a range — some of them $25 million wide.
• The reason is a cap, and it is the only one. House Rule XXV caps outside earned income at 15% of Executive Schedule Level II: $33,285 in 2025. You cannot check a cap against “$15,001–$50,000”. Ethics guidance requires the exact value for earned income expressly unlike every other interest on the same form.
• More lines have no ceiling than have a number. 6,001 asset rows sit in an open-ended top category that states a floor and stops — against 4,128 exact rows. A $51M holding and a $500M holding are the same row.
• The exact number is small. Median exact figure on the whole form: $7,399. The precision is spent on speaking fees and board stipends.
• The eight-figure income lines are arrival filings, not breaches — a first disclosure reports the year before office, when no cap applied. They are the last precise photograph before the ranges take over.

1. Everything you have read about congressional wealth is an estimate

There is a reason every net-worth figure you have seen for a member of Congress comes with a hedge. The disclosure form does not ask for amounts. It asks for categories: a holding is “$1,001–$15,000”, or “$500,001–$1,000,000”, or, at the top, “over $50,000,000” — a floor with nothing above it. Everyone who publishes a number, GovGreed included, takes the midpoint of each range and adds them up. It is the only thing you can do, and it is an estimate by construction.

That is usually explained as a privacy compromise, and as a compromise it is at least arguable. What is harder to explain is that the compromise stops. On the same form, in the same filing, one schedule asks for the dollar.

2. Data and methodology

Members file an annual Personal Financial Disclosure listing assets, liabilities, earned income and outside positions. We parsed 3,995 of them into structured rows — the machine-readable ones; 435 more exist only as scanned images and are excluded throughout. Coverage is 2018 to 2025.

For each schedule we asked one question: does the row carry an exact amount, or a bracket (a floor and a ceiling)? Nothing else. No midpoints, no sums, no net worth — this paper deliberately computes almost nothing, because the finding is in the shape of the form rather than in any total derived from it.

What we do not claim. This paper makes no compliance claim and tests no member against the cap. Doing so would require knowing which rows count as outside earned income under House Rule XXV — pensions, retirement distributions and investment income do not — and whether the filer was a sitting member during the year reported. Neither is determinable from these rows, and section 5 shows exactly why the naive version of that test would be wrong.

3. The wall

We expected a gradient: some schedules more precise than others, a spectrum of rigour. Table 1 is what the form actually does.

Read this table carefully — the two sides are not symmetrical evidence. Rows are from machine-readable annual Personal Financial Disclosures (parse_quality='parsed'), 2018–2025. The income row is a measurement: the schedule carries an exact-value field and bracket fields, and across 4,128 rows the bracket fields are used zero times. The liability and asset rows say “no such field” because the form offers no exact-value box on those schedules at all — so “none are exact” there is what the form provides, sourced to the ethics guidance in section 4, and is not a count we made. The 18,472 asset rows in neither column carry no value at all — unvalued holdings and none-reported lines. Verified July 15, 2026.

No schedule is mixed, and the earned-income side is the half that proves it. That schedule has somewhere to put a range — the fields are there, the same parser fills them 186,635 times elsewhere — and in 4,128 rows it never once needs them. The exactness is not a tendency. It is total.

The boundary between precision and vagueness runs exactly along the schedule line. That is the tell: the difference was designed, not accumulated.

Add the valued rows together and 4,128 of 190,763 carry a number: 2.2 percent.

4. Why: the one rule that needs a number

The exactness is not an accident and it is not our parser. Congressional ethics guidance states plainly that for earned income — unlike other financial interests — the exact value must be disclosed rather than a category of value. The form's own instructions carve it out.

The reason is that earned income is the one schedule someone is checking against a limit. House Rule XXV, created by the Ethics Reform Act of 1989 and effective from 1991, bars a member from having outside earned income above 15 percent of the annual rate of basic pay for Level II of the Executive Schedule. Level II was $221,900 on January 1, 2025, so the 2025 limit is $33,285. The House Ethics Committee republishes that figure every January.

Which produces the paper's actual claim, and it is a claim about architecture rather than about people. The form is not precise where money is large, or where conflict is likely, or where the public interest is greatest. It is precise where a rule exists that needs a number. There is exactly one such rule, and there is exactly one exact schedule. They are the same schedule.

Everywhere else the compromise holds — not because those numbers are more private, but because nothing downstream would do anything with them.

5. The eight-figure lines, and why they are not the story

If the cap is $33,285, an obvious test suggests itself: find the exact income rows above it. It is the wrong test, and the rows show why.

Largest exact figures in the earned-income schedule, all marked Self on the filing. Filing year is the year the disclosure covers. These are arrival filings — the first disclosure a candidate or incoming member submits reports the calendar year before they took office. Nothing here is a breach of any cap, and this paper makes no compliance claim. Verified July 15, 2026.

Every one of them is somebody's first disclosure. A candidate's or incoming member's opening filing reports the calendar year before they took the oath — a year in which House Rule XXV did not apply to them, because they were not yet a member. The fourth row says so out loud: the filer described the money as an exercise of options “received prior to entering Senate.”

There is something quietly sad in Table 2 anyway. These are the last precise photographs of these people's finances. From the filing after this one, the same person's money becomes a set of ranges — and the more of it there is, the wider the ranges get, until the top one stops having a ceiling.

6. Limitations and caveats

Machine-readable filings only. 3,995 parsed; 435 exist as scanned images and are excluded. If scanned filings differ systematically in schedule composition, the shares shift — the wall (100% / 0%) would not, since it is a property of the form.

“Exact” describes the field, not the truth. A filer writing an exact figure can still write the wrong one. This paper measures what the form demands, not whether filers comply.

The two halves of Table 1 are not symmetrical evidence, and v1.1 says so. The earned-income row is a real measurement: that schedule carries bracket fields as well as an exact-value field, and across 4,128 rows the bracket fields are used zero times — while the same parser fills bracket fields 186,635 times on the other schedules. The asset and liability rows are different. The form offers no exact-value box on those schedules, so “none of them are exact” is not a count we performed — it is what the form provides, and it rests on the ethics guidance cited in section 4 rather than on our data. v1.0 of this paper printed “0” in that column as though it had been measured. It had not been, and the column now reads “no such field.” The finding is unchanged; the evidence for each half of it is now labelled.

The 18,472 asset rows in neither column carry no value at all — unvalued holdings, none-reported lines, parse gaps. They are excluded from the 2.2 percent denominator rather than assumed into either side.

All 4,128 income rows are marked Self, and the raw filing text confirms the label rather than defaulting to it — the same parser records self / spouse / joint / dependent across 197,495 asset rows. We nonetheless do not claim these are all the earned income disclosed by these members: spouse earned income is reported differently, and its absence here may be a property of our parse rather than of the filings. The precision finding does not depend on it.

$33,285 is the 2025 figure. The limit is recomputed annually from Level II of the Executive Schedule; it was lower in earlier years of the window. Nothing in the argument depends on the exact value — only on a cap existing.

The legal claims are external. The exact-value requirement, House Rule XXV, the 15 percent formula, the $221,900 Level II rate, the $33,285 limit and the Ethics Reform Act of 1989 all come from published ethics guidance and House rules, not from our dataset. They are the load-bearing half of this paper and are cited as external throughout.

7. Conclusion

The standard defence of category reporting is privacy: a public servant should not have to publish their exact balance sheet to serve. It is a real argument, and this paper does not dispute it. It observes that Congress did not actually apply it.

A $7,399 speaking fee — the median exact figure on the entire form — receives no privacy protection whatsoever. It is published to the cent, because a rule needs to know whether it plus everything like it exceeds $33,285. A nine-figure portfolio receives an open-ended category, because no rule downstream is waiting for the number. Privacy is not the variable. Enforcement is.

Which makes the form an unusually honest document about its own authors. It is not that Congress cannot report precisely, or that precision is technically hard, or that the public cannot be trusted with it. The mechanism exists, is already law, and is complied with every year — on the one schedule where somebody attached a limit.

Six thousand and one lines have no ceiling. Four thousand one hundred and twenty-eight have a number. The gap between those two counts is not an accident of drafting. It is a map of what Congress decided to check.

Frequently asked

References & data sources

1. Congressional Personal Financial Disclosures — annual filings (House Clerk; Senate Office of Public Records). 3,995 machine-readable filings, 2018–2025: 4,128 earned-income rows, 197,495 asset rows, 7,612 liability rows. 435 scanned-only filings excluded.
2. Exact-value requirement for earned income — congressional ethics guidance states that for earned income, unlike other financial interests, the exact value must be disclosed rather than a category of value. Outside earned income above $200 is reportable.
3. House Rule XXV — limits outside earned income attributable to a calendar year to 15 percent of the annual rate of basic pay for Level II of the Executive Schedule. Created by the Ethics Reform Act of 1989, effective 1991, which also eliminated honoraria.
4. The 2025 limit — Level II of the Executive Schedule was $221,900 as of January 1, 2025; 15 percent of that is $33,285, the figure the House Committee on Ethics publishes annually.
5. Companion papers — GGR-WP-2026-16 (what the liability schedule does not ask), GGR-WP-2026-10 and GGR-WP-2026-08 (the bracket-midpoint arithmetic this paper declines to perform).
6. Image credit — Longworth House Office Building — Architect of the Capitol, public domain, via Wikimedia Commons.

Keep pulling the thread

Not financial advice. All data from public federal disclosures.