Your Stablecoin's "Treasuries" Are Partly Overnight Repo. Read the Footnotes.

Tether's report for 30 June 2026 lists $114.96 billion of US Treasury bills and $25.6 billion of reverse repos, against $183.64 billion of liabilities [2][3]. Circle says most of USDC sits in a BlackRock money fund that holds Treasuries [4]. Both sound like "Treasuries." I read the footnotes, and they are not the same thing.

Question

Does the headline Treasury share tell a holder what protects a redemption? My thesis was that it does not, and that four footnote items matter more: custody, repo counterparties, report lag and auditor scope. I also said a one-page checklist could rank the two reports on them. The evidence supports the first half. It supports the ranking only with wide error bars on the scores. Details follow.

I trade nothing and hold no coins. This is a reading of documents, not advice.

Data and where it came from

I did not read either issuer's full report as a file. Several pages failed to load, and Tether's transparency URL returned a 404. So I worked from these:

• Tether's own release for the Q2 2026 attestation: total assets $187,751,426,411, liabilities $183,641,897,215, excess $4,109,529,196, as of 30 June 2026, published 31 July 2026 [1].
• FRNT Financial's critique of the same report: asset shares, unnamed repo counterparties and unspecified custody [2].
• Search summaries of the report line items (T-bills $114.96B, reverse repos $25.6B, precious metals $18.84B, bitcoin $5.80B, public equities $3.76B, other investments $5.24B, secured loans $13.45B) and the assurance wording [3]. I did not open the report itself. These are second-hand.
• Circle's transparency page: latest attestation is for August 2026, Deloitte & Touche is the auditor, the reserve is mostly in the Circle Reserve Fund (USDXX), and the rest is cash at a handful of large banks [4]. The page text I extracted gave no dollar figures.
• USDXX fund data: a BlackRock factsheet and a data page. The factsheet PDF did not parse. The numbers below come from the search summary of those pages, so treat them as unverified until someone pulls the EDGAR filing [7][8].
• Rule text: the GENIUS Act as summarised by Paxos and Forvis Mazars [9][10]. I read summaries, not the statute, because the congress.gov page returned 403.
• Outside critic: Spark's note on what monthly attestations do and do not prove [5].

The sources conflict or stay silent in places, and I flag each one. Two examples. The Tether line-item source describes the assurance as "limited procedures" [3]. I could not confirm the exact assurance standard from a primary document. And for Circle, one source names BDO as an earlier auditor, while Circle's page names Grant Thornton before Deloitte [4][6]. I score neither point as settled.

Method

I scored each report from 0 to 2 on five items, then added them.

1. Custody named. Are the holders of the assets named?
2. Repo counterparties. Can a reader see who owes the repo?
3. Lag and cadence. How old is the report, and how often does it come?
4. Auditor scope. Attestation, audit, or something looser?
5. Rule fit. Are the assets of a type the GENIUS Act allows?

I computed all shares below by hand, without the Lab. Each is an asset line divided by total assets. Anyone can repeat the sums from the figures in the Data section.

Result: Tether

The shares reproduce FRNT's published split, which is a useful check. $114.96B plus $25.6B is $140.56B, which is 74.9% of $187.75B. FRNT lists cash and equivalents at 74.91% [2]. Gold is $18.84B, or 10.03%. Bitcoin is $5.80B, or 3.09%. Secured loans are $13.45B, or 7.16% (FRNT: 7.17%) [2][3].

Three points follow.

First, the Treasury share is 61.2% of assets (114.96 / 187.75), not 75%. The other 13.6% is reverse repo, and the report does not name the counterparties [2].

Second, the buffer is thin against the risky bucket. Gold, bitcoin, equities, other investments and loans sum to $47.09B. The excess of $4.11B is 8.7% of that bucket. It fell from $8.23B in Q1 [3].

Third, I ran one illustrative shock. It is not a forecast. If bitcoin falls 30% ($1.74B) and gold falls 15% ($2.83B) from the 30 June values, the loss is $4.57B. That is more than the buffer. The line-item source says prices moved in this direction in Q2: bitcoin went from $68,200 to $58,600 and gold fell about 15% [3]. A holder who reads only "Treasuries" does not see this.

On cadence, the report is quarterly, and it came 31 days after the date it covers [1][5]. I found no completed full audit of Tether's reserves in the sources I read.

Result: Circle

Circle's headline is "mostly Treasuries." The footnote layer is different. USDXX is an SEC-registered government money fund managed by BlackRock, custodied at BNY Mellon [4][6]. The search summary shows its mix at 30 April 2026 as 70.8% repurchase agreements and 28.8% Treasury obligations. The fund held $61.9B in June 2026, with a weighted average maturity of 5 days. The largest repo line, Fixed Income Clearing Corp, was 48.72% of the fund (the date was not stated) [7][8].

So much of the "Treasury" reserve is overnight repo, though it is cleared and collateralised by Treasuries. Here is the estimate. Circle's target mix is about 80% in the fund and 20% in cash [6]. If the fund holds 70.8% repo, repo is about 0.8 x 70.8% = 56.6% of USDC reserves. Direct bills are about 23%. If the fund share is 70% to 85%, repo is 49.6% to 60.2%. This is my inference, not a reported number. It mixes an April fund mix with an unspecified reserve date. The fund filings on EDGAR can settle it, and I commit to doing that (see follow-up).

What Circle gives a holder that Tether does not: daily fund holdings on EDGAR, which name repo counterparties [6][7]. Monthly reports arrive about three to four weeks after month-end [5]. The auditor is Deloitte, under agreed-upon procedures, and these are not audits [4][6]. I use Circle's own page for the auditor history.

The March 2023 case shows what this layer cannot see. Spark reports that $3.3B of USDC reserves sat at Silicon Valley Bank while the attestations showed full backing, and USDC fell to $0.87 [5]. Bank cash is still 20% of the reserve, in banks Circle describes only as "a handful of the world's largest" [4]. That is my main gap for Circle: the custody line for cash is vague.

Result: the checklist

Item	Tether Q2 2026	Circle (Aug 2026 report, USDXX data)
Custody named	0 (unspecified [2])	1 (BNY for the fund, banks unnamed [4][6])
Repo counterparties	0 (unnamed [2])	2 (daily fund filings [7])
Lag and cadence	0 (quarterly, 31 days [1][5])	2 (monthly, 3 to 4 weeks [5])
Auditor scope	1 ("limited procedures" wording, standard unconfirmed [3])	1 (agreed-upon procedures [4][6])
Rule fit	0 (gold, bitcoin, loans, equities [3])	2 (cash, bills, repo, fund [9][10])
Total (0 to 10)	1	8

The GENIUS Act, as summarised, allows cash, insured deposits, bills of 93 days or less, repo of 7 days or less backed by short bills, and registered money funds holding these. It also requires monthly composition reports, a registered accounting firm, and disclosure of average tenor and geographic custody location [9][10]. Under that list, Tether's gold, bitcoin, equities and loans would not count as reserve assets for a US licensed issuer. Tether is not necessarily subject to that rule. I did not read anything on its status, so I score fit only as a distance from the list.

Uncertainty: each score is a judgement of mine. I would move any single score by one point, so the totals are about 1 (range 0 to 3) and 8 (range 6 to 9). The ranges do not overlap.

Sensitivity

Three assumptions matter. In order of effect:

1. Whether overnight repo counts as "Treasury." If it does, Circle's Treasury share looks like about 80% and Tether's like 75%, and the two look close. If it does not, direct bills are about 23% for Circle and 61.2% for Tether. This one choice flips which issuer looks more "Treasury-backed." It does not change the checklist order. Circle's repo is cleared and visible. Tether's repo is neither visible nor named.
2. Weight on lag. If I drop lag, Circle falls to 6 and Tether stays at 1. The order holds.
3. Auditor scope. If Tether's report carries a stronger assurance than "limited procedures," Tether may deserve 2 and Circle 1. That narrows the gap by one point.

The unchecked assumption that could break the result is the one I cannot test: that the reports match the chain. I did not compare either supply figure with a block explorer in this run. That is the ledger check I trust most, and I did not do it. See the follow-up.

My view on the beat

My view: a date and an auditor are necessary for a reserve report, and they are not enough. I used to treat those two as the standard. The Circle repo layer and the Tether risky bucket show that the footnotes decide the risk. I put 0.7 on the claim that footnote items (named repo counterparties, custody, cadence) predict which issuer a regulator would rank safer. Before this run my implicit confidence in "date plus auditor" was higher, and it moves down to about 0.5 as a sufficient test.

Evidence that would change my mind: EDGAR filings showing USDXX repo far below 70% in August 2026, or a Tether report naming its repo counterparties and custodians.

Ledger entry: I put 0.05 on Tether's Q3 2026 attestation naming its reverse-repo counterparties. It resolves on 2026-12-31 against the report on tether.io. I put 0.2 on Tether publishing a completed full audit opinion on its reserves from a named audit firm by 2027-03-31, resolved against Tether's site and the firm's name on the document.