Federal Interest in Context — research working file

Compiled 2026-07-15. All figures pulled directly from primary sources listed at the bottom. Nothing in this file is from memory. Every number has a retrieval path.

0. TL;DR for the editor

The obvious angle is taken. "Interest is a record share of federal revenue, beating the 1991 peak" is already published, monthly, by the Peter G. Peterson Foundation, and has been fact-checked by Econofact because it was circulating on Reddit. Do not claim it.

The defensible original finding — and it is strong, because it lands twice, independently:

The "record" is an artefact of two arbitrary choices: which denominator you pick, and when your data series happens to start. Fix either one and the record disappears.

Kill shot #1 — the series starts in 1940 (§7a). FY2025's 18.53% is beaten by 30 of the 46 pre-1940 years. FY1878 hit 39.77%, FY1933 hit 34.52%, FY1923 hit 27.41%. Even after the most punitive possible gross-to-net adjustment, 22 of 46 still beat it. Today's ratio is the highest since 1940 — which is simply the year OMB's functional table begins. It is the highest in 92 years, not ever. (I verified the pre-1940 receipts against my own OMB file: FY1919–1933 match to the dollar, 10 of 10.)

Kill shot #2 — the denominator (§3). The record survives on exactly one denominator, by 0.09 of a percentage point, and reverses on OMB's own official on-budget concept by 1.89pp — twenty times the margin, with FY2025 ranking only 6th. Four defensible denominators split two-two:

The measures disagree for a reason that is itself the story: 1991's revenue base was more payroll-heavy (payroll = 37.54% of receipts in FY1991 vs 33.39% in FY2025), and 1991's Social Security trust funds were running a $52.2bn surplus that lent cash to the general fund, whereas FY2025's ran a $147.9bn deficit that drained it. Depending on which of those two facts you adjust for, you get a different champion.

The cleanest single sentence in the file: on OMB's own on-budget accounting — not our construction, OMB's — FY1991 remains the record at 28.21% against FY2025's 26.31%, and FY2025 is only 6th. The widely-repeated "record" exists solely on the total-receipts measure and solely by 0.09pp.

And the honest conclusion is not "the debt is fine." It is that the ranking is unknowable and the trend is undeniable. The un-contested fact that should anchor the close (§5): since FY2015, debt held by the public rose +142.3% while net interest rose +364.2% — the effective rate nearly doubled (×1.92). Roughly half the increase is borrowing more; half is paying more for it. That needs no denominator argument, no record claim, and no series-start date.

Novelty check: Econofact states the 1940 bound in passing; nobody has quantified what sits behind it (30 of 46 years, FY1878 at 39.77%). PGPF's tracker asserts "exceeding the previous high set in 1991" with no bound stated at all. The on-budget reversal and the two-part mechanism are unpublished. Rauh (2024) is the nearest prior art on the payroll-excluded idea — cite him.

1. Verifying the premise in the brief

Brief said: "$857bn in the first nine months of fiscal 2026, up 13 percent."

This is NET interest. Confirmed by reconstruction from MTS Table 5, June 2026:

• Net: 856.67 / 758.83 − 1 = +12.89% → rounds to the "up 13 percent" in the brief. ✅
• Gross: 1,052.02 / 920.97 − 1 = +14.23%.

So the brief's own numbers are net. Good — but note the trap below.

The brief's "deficit near $1.4tn" also checks out and is likewise a nine-month figure: FY2026 Oct–Jun outlays 5,517.918 − receipts 4,151.410 = 1,366.5bn. (FY2025's full-year deficit was $1.775tn.) So all three of the brief's numbers — $857bn, up 13%, ~$1.4tn — are internally consistent nine-month figures. None of them is annual. Any sentence pairing them with an annual number needs care.

The gross/net trap, stated precisely

• Gross interest = everything Treasury pays on all outstanding debt, including the interest it credits to itself on the $7.78tn of intragovernmental holdings (Social Security, Medicare, military retirement trust funds). That leg is an internal transfer. It is not a claim on taxpayers in the year it is paid.
• Net interest (budget function 900) = gross minus the trust-fund interest and minus other interest received. This is what actually leaves the government.
• FY2026 nine months: gross 1,052.0bn vs net 856.7bn. The gap is 195.3bn — about 23% larger than the net figure would suggest if you mixed them up.

Correction to a claim I initially made and then disproved by pulling the full series. My first read was "the gross/net gap is widening", based on the single YoY move (trust-fund interest +20.49% vs gross +14.23%; wedge 17.61% → 18.57% of gross). Over any horizon longer than one year that is flatly wrong. The wedge has collapsed:

42.01% → 18.57% in eleven years. The FY2025→FY2026 uptick is a one-year blip against a decade-long collapse. Mechanism: trust-fund balances have been roughly flat (Social Security is drawing down) while debt held by the public more than doubled, so a steadily larger share of gross interest is real money going to real creditors.

This is a genuine small finding and it is ours: the gross/net distinction used to be enormous and is becoming less so. Mixing up gross and net in FY2015 would have overstated the interest bill by 72.4% (318.2/184.6). Today it overstates by 22.8% (1,052.0/856.7). Still a serious error — but the trap is closing, and the reason it is closing is itself the debt story.

(Kept in the file deliberately as a record of a wrong turn caught by checking. Do not print the "widening" version.)

"$24bn a week": 856.67bn / 39 weeks (Oct 1 2025 – Jun 30 2026 ≈ 39.0 weeks) = $21.97bn/week net. On a gross basis it is 1,052.0 / 39 = $26.97bn/week. The "$24bn a week" in circulation matches neither cleanly — it is likely an annualised net figure (≈1,140bn / 52 = $21.9bn) or a rounded blend. Recommend not using the weekly framing unless we define it ourselves.

2. The ratio everyone quotes — interest as a share of receipts

(This section establishes the popular claim and reproduces it exactly. The actual finding is §3.)

2a. Total-receipts basis (the popular measure)

Source: OMB Historical Table 3.1 row "Net interest" ÷ Table 1.1 "Total Receipts". Both FY, both actuals through FY2025.

Full ranking, top 15 of 1940–2025:

Reference points requested in the brief:

• FY1995: 17.17% · FY2000: 11.01% · FY2015: 6.87% · FY2019: 10.83% · FY2023: 14.82%

The 1990s peak is FY1991 at 18.43%. FY2025 beat it by 0.10pp.

⚠️ 0.10pp is inside the noise. PGPF printed 18.6% in February 2026 and 18.5% now for the same concept. CRFB/Fortune print "19%". Reasonable people using slightly different receipts definitions get answers spanning ~0.5pp. Do not stake a headline on a 0.10pp gap. This is the single most important editorial caution in this file.

2b. Nine-month, like-for-like (Oct–Jun), FY2015–FY2026

Built from MTS so every year is the identical nine-month window. MTS API coverage starts 2015-03, so FY2015 is the earliest complete Oct–Jun window available from this source.

$bn. Net/Rcpts on the Oct–Jun window has gone 7.54% → 20.64% in eleven years, a 2.7x rise.

2c. CAN nine months be annualised honestly? — tested, answer: only with a stated adjustment

Naively annualising is wrong: the Oct–Jun ratio systematically overstates the full-year ratio, because Q4 (Jul–Sep) receipts are seasonally strong relative to interest. Measured, not assumed:

Mean gap +0.90pp, median +0.91pp, stdev 0.71pp. Last five years mean +0.41pp.

→ FY2026 nine-month 20.64% implies a full-year 19.7% (mean-gap) to 20.2% (last-5-gap). Either way it clears both FY1991 (18.43%) and FY2025 (18.53%) on the total-receipts basis.

Honest phrasing for the piece: "running at 20.6% through nine months, against 18.9% in the same nine months of FY2025" — a like-for-like window, no annualisation required. If we must project, state the +0.9pp seasonal adjustment and the 0.71pp stdev.

3. KILL SHOT #2 — the record is denominator-dependent

(§3a–b = our first construction. §3c = the strongest version, using OMB's own concept. §3d = the counter-argument against our own finding, which partly succeeds. Read all four. Kill shot #1 — the pre-1940 series — is at §7a.)

3a–b. Net interest vs receipts excluding dedicated payroll taxes

Rationale for the alternative denominator: net interest is paid to the public, out of the general fund. Payroll taxes (OASDI, HI, unemployment, other retirement) are legally dedicated to trust funds. Since the net-interest concept already nets out what the trust funds receive, the conceptually matched denominator is receipts excluding those dedicated payroll taxes. This is the revenue that actually stands behind the interest bill.

Source: OMB Table 2.1 "Social Insurance and Retirement Receipts" subtracted from total receipts. Consistency check passed: Table 2.1 components sum to Table 1.1 totals to within $1m in every year tested (FY1991 diff −0.001bn; FY2015/2024/2025 diff 0.000bn).

Top 12, net interest ÷ non-payroll receipts, 1940–2025:

On this basis FY2025 is 5th, not 1st. FY1991 still leads by 1.70pp.

Why the two measures disagree — the mechanism

1991's revenue base was more payroll-heavy than today's (37.5% vs 33.4%) — the post-1983 Greenspan-reform Social Security surpluses were at full flood. So removing payroll taxes shrinks 1991's denominator proportionally more, pushing its ratio up more. Today's base leans harder on individual income tax. The "record" in the popular measure is partly a story about the changing composition of federal revenue, not purely about the interest bill.

Nine-month, non-payroll basis

9mo-vs-full-FY gap on this basis: mean +1.75pp, last-5 +0.43pp. → full-FY2026 estimate 29.20% – 30.52%, against FY1991's 29.51%.

FY2026 is a coin-flip against 1991 on the tougher measure, at the exact moment the popular measure declares a decisive record. That tension is the piece.

3c. The strongest version — OMB's OWN on-budget concept

This is better than my §3 construction because it is not my construction at all. OMB publishes net interest split on-budget/off-budget (Table 3.1, rows under "Net interest") and receipts split on-budget/off-budget (Table 1.1, cols 5 and 8). Using both consistently requires no analytical choice by us whatsoever.

Note the sign logic: off-budget net interest is negative (the Social Security trust funds receive interest), so on-budget net interest exceeds total net interest. FY1991: on-budget NI 214.7 vs total NI 194.4. This is exactly the gross/net issue reappearing at the on-budget boundary.

Top 8, on-budget net interest ÷ on-budget receipts, 1940–2025:

On OMB's own official on-budget accounting, FY2025 ranks 6th, and 1991–1992 still lead. This is the single most damaging fact for the "record" headline, and it requires no cleverness from us — only that both numerator and denominator be taken from the same budget concept, which the popular measure fails to do (it puts a total net-interest numerator over a total receipts denominator, which is internally fine, but is not the only internally-fine choice).

⚠️ Cannot be extended to FY2026. MTS does not publish an on-budget/off-budget split of net interest, and OMB's FY2026 column is an estimate. The on-budget fork is a FY2025-vs-FY1991 actuals argument only. That is sufficient — it is the year the record was declared.

3d. The honest counter-argument to my own §3 finding (tested, and it bites)

Objection: in 1991 the Social Security trust funds ran a large surplus and lent the excess to the general fund by buying Treasuries. So 1991's general fund had more cash than "non-payroll receipts" implies. Today the trust funds run a deficit and are redeeming, so today's general fund must find extra cash. If so, my §3 denominator over-penalises 1991 and under-penalises 2025.

Tested against OMB Table 1.1 off-budget surplus — the objection is correct:

Adjusting for it (denominator #4 in §0) flips the answer back to FY2025, 29.04% vs 27.34%.

⚠️ #4 is the roughest of the four and I would not lead with it. It is internally inconsistent: it subtracts all social-insurance receipts (including Medicare HI and unemployment, which are on-budget) but adds back only the off-budget (OASDI + Postal) surplus. It is directionally informative, not a clean concept. Its job is to show that §3's finding is itself denominator- sensitive — which is the whole thesis, honestly applied to my own construction.

Where this lands. Three defensible measures say 1991, one says 2025, one says 2025 by a hair, and my own best alternative can be flipped by a further adjustment. The intellectually honest conclusion is not "1991 still holds the record." It is: FY1991 and FY2025 are effectively tied, and every "record" claim in either direction is an artefact of denominator choice. That is a real, defensible, publishable finding — and it is more robust than the thing it corrects.

4. Angle 2 — interest vs every other budget function (FY2025 actuals)

OMB Table 3.1, FY2025 column, $bn:

Verdict: weak as a standalone angle. "Interest > defense" is the cliché the brief wanted us to beat, and the only step past it is "interest is 4th, within $8.4bn of Health and $26.6bn of Medicare" — which is true but is a near-miss, not a fact. It is a decent supporting paragraph (interest is on the verge of becoming the #2 line item in the entire federal government) but it cannot carry the piece.

⚠️ We cannot rank FY2026 by function. MTS Table 5 is organised by agency, not by budget function, and OMB's FY2026 column is an estimate, not an actual. So the FY2025 ranking above is the latest actual functional ranking available, full stop. For colour only, FY2026 Oct–Jun agency outlays ($bn): HHS 1,474.3 · Social Security Administration 1,293.4 · net interest 856.7 · Defense–Military 678.7 · Veterans Affairs 304.9. Do not present these as function rankings — HHS bundles Medicare, Medicaid and Health, so it is not comparable to the table above. The only clean FY2026 statement is that net interest (856.7) exceeds DoD Military Programs (678.7) by 26%.

5. Angle 3 — the compounding / rate effect

Debt held by the public at Jun 30 (Treasury Debt to the Penny), vs Oct–Jun net interest:

Decomposition: interest +364.2% = debt effect +142.3%, compounded with a rate effect of +91.6%. (1.6422 × 1.9160 = 4.642 → +364.2%. ✅)

So slightly under half the growth in the interest bill is borrowing more; slightly over half is paying more for it. The effective rate on publicly held debt has essentially doubled.

Debt at Jun 30 2026: public $31.681tn, intragovernmental $7.781tn, total $39.462tn.

Verdict: solid, and the least-worked of the four angles. Works well as the mechanism section supporting the main finding. Caveat: the "effective rate" here is a crude ratio (nine-month interest ÷ point-in-time debt stock), not a true weighted-average coupon. Treasury publishes an actual average interest rate series; if we print a rate number we should either use theirs or label ours explicitly as an implied ratio.

6. Angle 4 — per household

Dropped. Two reasons:

1. Census Table HH-1 (https://www2.census.gov/programs-surveys/demo/tables/families/time-series/households/hh1.xlsx) would not retrieve cleanly in this session (timeouts, then a non-xlsx payload), so I have no verified household count and will not use one from memory.
2. It is the most-worked framing in the genre and adds nothing the ratio does not already say.

If we ever want it: the honest denominator must be stated (households ≠ taxpayers ≠ people), and net interest is not actually billed to households, so the framing is rhetorical rather than analytical. Recommend against.

7. What the data CANNOT settle

1. Pre-1940 is now RESOLVED, not a caveat — it is finding #1 (§7a). OMB Table 3.1 begins in FY1940, and that start date is doing all the work in every "record" claim, PGPF's included. 30 of 46 pre-1940 years beat FY2025; the peak is FY1878 at 39.77%. What the data still cannot settle: the Civil War-era rows are one-source (Historical Statistics only — I could not corroborate them against OMB, whose annual rows start 1901), and pre-1866 was never examined, so the true all-time peak is unknown and may sit in the 1790s. Also, pre-1940 is gross interest under administrative-budget accounting — a different concept from modern net interest, though the wedge was near 1.0 then and the finding survives even the max wedge. Licensed: "highest since 1940" / "since the Great Depression" / "in 92 years". Forbidden: "ever", "in history", "all-time".
2. The 0.09pp margin (FY2025 vs FY1991) is not robust. It is smaller than the spread between credible publishers measuring the same concept (18.5 / 18.6 / 19). A finding that flips on a decimal is not a finding. This cuts against the popular claim — which is our thesis — but it also means we cannot claim the reverse on the total-receipts measure either.
3. Fiscal vs calendar year. Everything above is fiscal year (Oct–Sep). FRED's quarterly NIPA series (A091RC1Q027SBEA) is calendar-year and uses BEA national-accounts concepts, which do not equal the budget-function net interest. Do not mix them. (FRED CSV endpoints timed out repeatedly in this session and were not used for any number in this file.)
4. Nominal only. Nothing here is inflation-adjusted. The ratios are unit-free so this does not contaminate them, but any dollar figure quoted across decades ($194bn in 1991 vs $970bn in 2025) is nominal and should not be compared without saying so. The ratio is the inflation-proof comparison — that is precisely why it is the right frame.
5. Nine months is not a year. Quantified in §2c: the Oct–Jun ratio runs ~0.9pp hot vs the full year, stdev 0.71pp. FY2026 Q4 is unobserved. A September receipts surprise or a rate move changes the landing spot.
6. The non-payroll denominator (§3) is an analytical choice, not an official concept. It is defensible (net interest is a general-fund obligation; payroll taxes are dedicated) but it is our construction, and §3d shows it can be flipped by a further defensible adjustment. Under unified-budget logic, trust-fund surpluses are lent to the general fund and the separation is partly notional. We must show all the measures and let the reader see the fork. Presenting only the non-payroll measure would be exactly the sin we are accusing others of. The on-budget measure (§3c) does not have this problem — it is OMB's own concept — which is why it should carry the piece.
7. Which year is "the record" is not settleable, and that is the finding. The data cannot tell you whether 1991 or 2025 was worse, because the answer depends on a denominator choice that is genuinely contested and on which reasonable analysts differ. Anyone claiming otherwise — in either direction, including us — is overstating what the numbers support. The piece must assert the tie, not a new champion.
8. The on-budget fork cannot be run for FY2026. MTS publishes no on-budget/off-budget split of net interest, and OMB's FY2026 column is an estimate. §3c is a FY2025-vs-FY1991 actuals argument.
9. "Effective rate" in §5 is an implied ratio (nine-month interest ÷ point-in-time debt stock), not Treasury's published average interest rate on the debt. Label it as such or use Treasury's.
10. FY2026 and FY2027 columns in OMB tables are estimates, not actuals. Every number I have used from OMB is an actual (FY2025 and earlier); all FY2026 figures come from MTS actuals only.
11. MTS is subject to revision, which matters specifically because finding #1 rests on 0.09pp.
12. Agency ≠ function (see §4). The FY2026 outlay figures I have are agency-based and cannot be ranked against the FY2025 functional table.

7a. Pre-1940 — RESOLVED. This is KILL SHOT #1 and the strongest thing in the file.

(Filed here because it began as a caveat to §7. It is not a caveat. It is the lede.)

The "record" is an artefact of when the data series starts. FY2025's 18.53% is beaten by 30 of the 46 pre-1940 years for which interest and receipts can both be sourced.

FY1919–1940, gross interest ÷ receipts:

Three different mechanisms, not one: Civil War debt (1875–82), WWI debt on a small pre-modern revenue base (1922–25), and the Depression collapse in receipts (1932–33 — interest actually fell from $1,056m in 1923 to $612m in 1931; the spike is the denominator, not the numerator).

I independently verified this against the OMB file I hold locally

Cross-checking the agent's receipts against OMB Table 1.1: FY1919–1933 match to the dollar, 10 of 10 tested (1919, 1920, 1921, 1922, 1923, 1929, 1931, 1932, 1933 + 1940). FY1934 and FY1939 do not match (OMB 2,955 vs 3,015; OMB 6,295 vs 4,979) — which confirms the researcher's own stated caveat that Historical Statistics' administrative-budget basis diverges from OMB from 1934 as trust funds appear. The divergence begins exactly where they said it would. Recomputing from OMB's receipts: FY1923 = 27.41%, FY1932 = 31.13%, FY1933 = 34.50%. The peak window reconciles independently.

The gross/net caveat — quantified, and the verdict survives it

Pre-1940 figures are gross interest; there is no pre-1940 net-interest concept because trust funds barely existed (civil service retirement was $8–21m/year through 1932; Social Security began collecting 1937). So gross ≈ net in the peak years. Bounding it with the largest gross/net wedge ever observed on the 1940–53 overlap (1.261×, FY1953):

Even under the most punitive adjustment the finding holds.

⚠️ Data-quality warning — do not lift these from the PDF text layer

The FRASER PDF's OCR text layer is materially wrong: it returns 699,277 for FY1932 against a true 599,277 — a 17% error that would have falsely made 1932 the interwar peak instead of 1933. Three such errors were found (FY1923, 1931, 1932) and corrected by rendering pages as images and checksumming components against published totals, then confirmed against Treasury's independent Table 4. Any re-check must use the page images, not the text layer.

What remains unverified

• FY1878/1879/1875 are one-source (Historical Statistics only; OMB Table 1.1's annual rows start 1901, so I could not corroborate them locally). Three Civil War-era rows (1875, 1881, 1882) fail the researcher's checksum by small amounts; 1878 and 1879 both reconcile exactly. Treat the 1866–1890 block as strong but uncorroborated.
• FY1934–1939 interest is HS-only (Treasury's FY1933 report ends at 1933). None are near the peak.
• Pre-1866 not examined at all. Revolutionary War debt in the 1790s could plausibly be higher still. Make no claim about it.

What this licenses us to write

✅ "highest since 1940" · ✅ "highest since the Great Depression" · ✅ "highest in ~92 years" ❌ never "highest ever" / "in history" / "all-time record"

Sources: Historical Statistics of the United States, Colonial Times to 1970 (Census, 1975), Vol. 2, Series Y 457–465 col. 461 "Interest on the public debt," pp. 1114–15, and Series Y 352 "Federal Government Receipts—Administrative Budget: 1789 to 1939," p. 1106 — https://fraser.stlouisfed.org/files/docs/publications/histstatus/hstat1970_cen_1975_v2.pdf Independent confirmation of the 1916–33 interest series: Annual Report of the Secretary of the Treasury, FY1933, Table 4, p. 301 — https://fraser.stlouisfed.org/files/docs/publications/treasar/AR_TREASURY_1933.pdf (Treasury states FY1923 interest as $1,055,923,689.61.)

8. Prior published analysis — CHECKED, AND IT IS BAD NEWS FOR THE OBVIOUS ANGLE

The headline comparison is already published, repeatedly, by the most-cited source in the field.

• PGPF, Interest Costs on the National Debt tracker, updated monthly, currently through June 2026: "As a share of federal revenues, federal interest payments rose to 18.5 percent by the end of last year, exceeding the previous high set in 1991." https://www.pgpf.org/programs-and-projects/fiscal-policy/monthly-interest-tracker-national-debt/ — I fetched this myself and confirmed the language and the June 2026 data cutoff. It is our thesis, sentence for sentence, on a monthly refresh.
• PGPF, updated 2026-02-12: "would rise to 18.6 percent this year, above the previous high set in 1991." https://www.pgpf.org/article/any-way-you-look-at-it-interest-costs-on-the-national-debt-will-soon-be-at-an-all-time-high/
• Econofact / Gigafact fact brief, 2026-06-09: "In fiscal year 2025, 18.5% of the federal government's revenue went to paying interest on the national debt, a record since the start of the data series in 1940." Previous high 18.4% in 1991. https://econofact.org/factbrief/fact-check-did-interest-payments-on-the-federal-debt-represent-a-record-share-of-revenue-in-fy2025 — Note what this is: a fact-check of a claim already circulating on r/economy. The claim is not just published, it is popular enough to need debunking infrastructure.
• Joshua Rauh (Hoover), Liberty Lens Econ, 2024-10-08: had the receipts framing and the 1991 anchor, and a payroll-excluded variant (he cites 25–28%). https://libertylensecon.substack.com/p/playing-with-fire-the-interest-coverage — This is the closest prior art to my §3 finding. He gets to the same neighbourhood. What I have that he does not: the full 1940–2025 ranking on the non-payroll basis, the finding that 1991 still leads on it, and the payroll-share mechanism that explains why. That is a real increment, but it is an increment on his idea, not a new idea.
• CRFB, 2026-05-21, amplified by Fortune 2026-05-27 ("Interest on the national debt is eating a record 19% of federal revenue"): receipts denominator, record claim, no 1991 anchor. https://www.crfb.org/blogs/rising-interest-rates-are-exploding-debt https://fortune.com/2026/05/27/national-debt-interest-payments-30-percent-revenue-bond-yields-crfb/
• EPIC for America, Interest Spending Tracker, 2026-01-21: published the identical fiscal-year-to-date construction ("Net interest outlays were equivalent to 22.1% of total revenues through Q1 of FY 2026"), including the prior-year same-period comparison. Stopped updating 2026-06-01. https://epicforamerica.org/federal-budget/interest-spending-tracker-q1-of-fy-2026/
• CBO Monthly Budget Review, June 2026 hands over both inputs to the nine-month ratio directly. https://www.cbo.gov/publication/61982
• Also using the receipts denominator in some form: American Action Forum, Heritage, CNN, Fox Business.

Not found anywhere: the specific Oct–Jun FY2026 ratio (20.64% vs 18.93%), and the full non-payroll historical ranking showing 1991 still on top. Those are thin reeds — the former is one division away from a public CBO document, and the latter is an extension of Rauh.

9. Recommended structure (my honest read)

Do not write the "record share of revenue" piece. PGPF publishes it monthly, Fortune ran it in May, and Econofact already fact-checked it. We would be last to a crowded party and it would look like aggregation, which is exactly the line this outlet is not supposed to cross.

The piece I would write instead — "The record that isn't":

1. Lede — the number everyone is citing (18.5% of every tax dollar, a record, beats 1991) is true on its own terms and is carrying far more weight than it can bear. It is a record only because the data series starts in 1940.
2. Kill shot #1, the big one (§7a) — go back before 1940 and 30 of 46 years beat it. FY1878: 39.77%. FY1933: 34.52%. FY1923: 27.41%. Today is the highest in 92 years, not ever. Show the three distinct mechanisms — Civil War debt, WWI debt on a tiny revenue base, and the Depression collapse in receipts (interest fell through the 1920s; the 1933 spike is the denominator). Concede the gross/net issue immediately and show the finding survives the max-ever wedge (22 of 46).
3. Establish we did the arithmetic, not the aggregation — the full 1940–2025 ranking, FY2025 18.53% vs FY1991 18.43%. Then the margin point: 0.09pp is thinner than the disagreement among the people publishing the claim (PGPF printed 18.6 in February and 18.5 in June; CRFB and Fortune say 19). A finding that flips on a decimal is not a finding.
4. Kill shot #2 — and make it OMB's own concept, not ours (§3c). On-budget net interest ÷ on-budget receipts: FY1991 28.21%, FY2025 26.31%. FY2025 is 6th. Strongest card of the two because we invented nothing; we only insisted numerator and denominator come from the same budget concept.
5. The mechanism — payroll taxes were 37.5% of receipts in 1991 vs 33.4% now; and 1991's Social Security ran a +$52.2bn surplus lending to the general fund while 2025's ran a −$147.9bn deficit draining it. The "record" is substantially a story about how the revenue base and the trust funds changed.
6. Turn the knife on ourselves (§3d) — adjust for that trust-fund flow and the answer flips back to 2025. Show this. It is the difference between an argument and a hit job, and it earns the conclusion.
7. Land it honestly — the conclusion is not "1991 was worse" and emphatically not "the debt is fine." It is: the ranking is unknowable; the trend is undeniable.
8. Close on what is unambiguously true (§5) — the rate effect. Debt +142.3%, interest +364.2%, effective rate ×1.92 since FY2015. Half the increase is borrowing more, half is paying more for it. This is the load-bearing un-contested fact. It needs no denominator, no record claim, no series-start date — and it is what stops the piece reading as debt minimisation.
9. Method box — gross vs net and why we use net (§1), the $195.3bn gap and why mixing them inflates the story by ~23%, FY not CY, nominal, the +0.9pp nine-month seasonal adjustment with its 0.71pp stdev, the "since 1940" bound, and the honest admission that pre-1866 is unexamined so the true all-time peak is unknown.

Bonus sidebar available (§1): the gross/net wedge has collapsed from 42.0% of gross interest in FY2015 to 18.6% today. Conflating the two used to overstate the bill by 72%; now it overstates by 23%. Nobody has published this, it is a clean arithmetic fact from one MTS pull, and it is a natural companion to the method box — it makes the gross/net lecture interesting rather than housekeeping.

Why this clears the novelty bar: we are not reporting the record, we are auditing it. The contributions nobody has published: (a) the pre-1940 series showing 30 of 46 years beat FY2025 — Econofact mentions the 1940 bound in one clause but nobody has quantified what is behind it, and PGPF's tracker states no bound at all; (b) the 0.09pp margin-of-error point set against the publishers' own spread; (c) the on-budget ranking showing FY2025 is 6th on OMB's own concept; (d) the two-part mechanism (payroll share + trust-fund flow direction); (e) the self-refutation in §3d. Rauh got nearest — he had the payroll-excluded idea in 2024 — but he did not run the historical ranking, did not find the on-budget reversal, and did not make the margin argument. Cite him. We are extending him, not discovering it, and the piece should say so.

(a) is the strongest and most defensible thing in this file. It is archival work against a primary source, it required catching a 17% OCR error that would have produced the wrong peak year, and it converts a claim everyone is repeating into a claim about a spreadsheet's start date.

Risks to accept going in: this is a contrarian methodological piece against PGPF, CRFB and a fact-check that all say "record." We would be arguing the alarming number is softer than reported. That is defensible and differentiated, but it must be flawlessly sourced and must not read as debt minimisation — the rate-effect close (§5) is what stops it doing so, because it concedes the underlying trend is real and severe.

If PK does not want a corrective piece: the honest fallback is the rate-effect decomposition (§5) as a standalone explainer. Clean, arithmetic, unowned, no record claim. But it is a modest piece, and I would rather say that plainly than dress it up.

What I would NOT write: the straight "interest is a record share of revenue" piece. PGPF publishes it monthly with a June 2026 cutoff, Fortune ran it in May, Econofact fact-checked it in June. We would be last to it, and it would read as aggregation — which is the line this outlet is specifically not supposed to cross.

10. Sources — exact retrieval paths

Treasury Fiscal Data API (no key):

• MTS Table 5, June 2026, all rows: https://api.fiscaldata.treasury.gov/services/api/fiscal_service/v1/accounting/mts/mts_table_5?filter=record_date:gte:2026-06-01,record_date:lte:2026-06-30&page[size]=1000 Lines used: 4176 Total—Interest on Treasury Debt Securities (Gross); 5680 Total—Interest Received by Trust Funds; 5576 Other Interest; 5691 Total Outlays.
• MTS Table 5 time series (interest + outlays, 2015-03 → 2026-06): https://api.fiscaldata.treasury.gov/services/api/fiscal_service/v1/accounting/mts/mts_table_5?filter=line_code_nbr:in:(4176,5680,5691)&fields=record_date,line_code_nbr,classification_desc,current_fytd_net_outly_amt&sort=record_date&page[size]=500
• MTS Table 4 receipts time series. Lines: 830 Total—Receipts; 286 Total—Social Insurance and Retirement Receipts: https://api.fiscaldata.treasury.gov/services/api/fiscal_service/v1/accounting/mts/mts_table_4?filter=line_code_nbr:in:(830,286)&fields=record_date,line_code_nbr,current_fytd_net_rcpt_amt&sort=record_date&page[size]=400
• Debt to the Penny, every Jun 30: https://api.fiscaldata.treasury.gov/services/api/fiscal_service/v2/accounting/od/debt_to_penny?filter=record_calendar_month:eq:06,record_calendar_day:eq:30&fields=record_date,debt_held_public_amt,intragov_hold_amt,tot_pub_debt_out_amt&sort=record_date&page[size]=60
• MTS API coverage begins 2015-03-31 (verified by sorting ascending). This is why the like-for-like nine-month series starts at FY2015.

OMB Historical Tables (FY2027 edition, published April 2026):

• Table 1.1 Summary of Receipts, Outlays, Surpluses/Deficits: https://www.whitehouse.gov/wp-content/uploads/2026/04/hist01z1_fy2027.xlsx
• Table 2.1 Receipts by Source (used for payroll/social-insurance receipts): https://www.whitehouse.gov/wp-content/uploads/2026/04/hist02z1_fy2027.xlsx
• Table 3.1 Outlays by Superfunction and Function (row "Net interest", row 21 0-indexed): https://www.whitehouse.gov/wp-content/uploads/2026/04/hist03z1_fy2027.xlsx
• Index: https://www.whitehouse.gov/omb/budget/historical-tables/

Pre-1940 (§7a) — archival, via FRASER:

• Historical Statistics of the United States, Colonial Times to 1970 (Census, 1975), Vol. 2: https://fraser.stlouisfed.org/files/docs/publications/histstatus/hstat1970_cen_1975_v2.pdf — Interest: Series Y 457–465, col. 461 "Interest on the public debt," pp. 1114–15 — Receipts: Series Y 352 "Federal Government Receipts—Administrative Budget: 1789 to 1939," p. 1106
• Annual Report of the Secretary of the Treasury, FY1933, Table 4, p. 301 (independent confirmation of the 1916–33 interest series; states FY1923 interest as $1,055,923,689.61): https://fraser.stlouisfed.org/files/docs/publications/treasar/AR_TREASURY_1933.pdf
• ⚠️ Use the page images, not the OCR text layer — the text layer returns 699,277 for FY1932 against a true 599,277 (17% error, would have moved the interwar peak from 1933 to 1932).
• ✅ My independent cross-check: the pre-1940 receipts reconcile against OMB Table 1.1 (held locally) to the dollar for 1919–1933, 10 of 10 tested. FY1934 and FY1939 diverge (OMB 2,955 vs 3,015; 6,295 vs 4,979) exactly as predicted by the administrative-vs-unified budget shift from 1934.

Not used / failed:

• FRED CSV endpoints (fredgraph.csv?id=FYFR, FYOINT, GFDEBTN) — repeated timeouts (curl exit 28). No FRED number appears in this file.
• Census Table HH-1 — timeout, then non-xlsx payload. No household count appears in this file.

Local working copies: C:\Users\W11\Documents\Claude\Projects\MoneyAndWorld\research\data\ (mts5_jun2026.json, mts5_series.json, mts4_series.json, mts4_si.json, debt_june.json, hist01z1_fy2027.xlsx, hist02z1.xlsx, hist03z1_fy2027.xlsx, omb_series.json)

11. Arithmetic audit trail

Net interest FY26 9mo  = 1,052,019,385,978.85 − 195,345,509,160.97 − 4,365,479.58
                       = 856,669,511,338.30
Net interest FY25 9mo  =   920,965,036,885.91 − 162,130,919,161.64 − 5,856,566.25
                       = 758,828,261,157.02
YoY net                = 856.6695 / 758.8283 − 1 = +12.894%   → "up 13%" ✅
YoY gross              = 1052.0194 / 920.9650 − 1 = +14.230%
Trust-fund int YoY     = 195.3455 / 162.1309 − 1 = +20.487%
Trust int / gross      = 133.71/318.23 = 42.014% (FY15)  →  195.35/1052.02 = 18.569% (FY26)
                         COLLAPSING, not widening. FY25→FY26 (17.606→18.569) is a 1yr blip.
Gross/net overstatement = 318.23/184.55 = 1.724 (FY15, +72.4%)
                        = 1052.02/856.67 = 1.228 (FY26, +22.8%)

Raw cells, verified directly against the sheets (all $ millions):
  OMB T3.1 col 53 = FY1991: National Defense 273,285 | Net interest 194,448 | Total outlays 1,324,226
  OMB T3.1 col 87 = FY2025: National Defense 916,140 | Net interest 970,065 | Total outlays 7,011,105
  OMB T1.1 FY1991: Receipts 1,054,988 | Outlays 1,324,226 | Deficit −269,238
  OMB T1.1 FY2025: Receipts 5,236,421 | Outlays 7,011,105 | Deficit −1,774,684

FY25 total-receipts basis   =  970.065 / 5,236.421 = 18.526%
FY91 total-receipts basis   =  194.448 / 1,054.988 = 18.431%    margin = 0.095pp
FY25 non-payroll basis      =  970.065 / 3,488.1   = 27.812%
FY91 non-payroll basis      =  194.448 /   659.0   = 29.506%    1991 leads by 1.694pp
Payroll share FY91          =    396.0 / 1,054.988 = 37.536%
Payroll share FY25          =  1,748.3 / 5,236.421 = 33.388%

Deficit cross-check on the brief's "near $1.4tn":
  FY26 Oct–Jun outlays 5,517.918 − receipts 4,151.410 = 1,366.508bn ✅
  (This is the NINE-MONTH deficit, not an annual figure. FY2025 full-year deficit was
   $1.775tn. The brief's $1.4tn and $857bn are both nine-month figures — consistent
   with each other, but neither is annual. Do not pair either with an annual number.)

FY26 9mo total-receipts     =  856.6695 / 4,151.410 = 20.636%
FY25 9mo total-receipts     =  758.8283 / 4,008.146 = 18.933%
FY26 9mo non-payroll        =  856.6695 / 2,767.731 = 30.952%
FY25 9mo non-payroll        =  758.8283 / 2,658.846 = 28.539%

THE FOUR-DENOMINATOR FORK (FY1991 vs FY2025, $bn):
  #1 total receipts        194.448/1,054.988 = 18.431%  |    970.065/5,236.421 = 18.526%  → 2025 +0.095pp
  #2 non-payroll receipts  194.448/  659.0   = 29.506%  |    970.065/3,488.1   = 27.812%  → 1991 +1.694pp
  #3 on-budget/on-budget   214.670/  761.103 = 28.206%  |  1,040.091/3,952.685 = 26.314%  → 1991 +1.892pp
  #4 non-payroll+OB surp   194.448/(659.0+52.198)=27.339% | 970.065/(3,488.1−147.869)=29.042% → 2025 +1.703pp

  #3 raw cells, verified directly ($m):
     OMB T3.1 FY1991: Net interest 194,448 | (On-budget) 214,670 | (Off-budget) −20,222
     OMB T3.1 FY2025: Net interest 970,065 | (On-budget) 1,040,091 | (Off-budget) −70,026
     OMB T1.1 FY1991: total 1,054,988 | on-budget 761,103 | off-budget 293,884 | off-budget surplus +52,198
     OMB T1.1 FY2025: total 5,236,421 | on-budget 3,952,685 | off-budget 1,283,736 | off-budget surplus −147,869
     cross-checks: 194,448 + 20,222 = 214,670 ✅ | 970,065 + 70,026 = 1,040,091 ✅
                   5,236,421 − 1,283,736 = 3,952,685 ✅
  #3 sign note: off-budget net interest is NEGATIVE (trust funds RECEIVE interest),
                so on-budget NI (214.670) > total NI (194.448) in FY1991. Off-budget leg = −20.222.
  #4 inconsistency: subtracts ALL social-insurance receipts but adds back only the
                OFF-budget (OASDI+Postal) surplus. Directional only — do not lead with it.

  Off-budget surplus (OMB T1.1 col 10): FY1991 +52.2 ; FY2025 −147.9

Rate decomposition FY15→FY26 (Oct–Jun net int; public debt at Jun 30):
  debt   13,076.4 → 31,681.3   = ×2.4227  (+142.27%)
  int       184.55 →   856.67  = ×4.6418  (+364.18%)
  rate    ×4.6418 / ×2.4227    = ×1.9160  (+91.60%)
  check:  2.4227 × 1.9160 = 4.6419 ✅
  implied 9mo eff. rate  184.55/13,076.4 = 1.4113%  →  856.67/31,681.3 = 2.7040%