Dollar-funding capacity has a structural warning, not a confirmed funding squeeze. Seiche’s board reads 45.3, STRAIN, but its largest contribution is a dated reserve-path calculation: the $210B Treasury auction settlement scheduled for 2026-08-25. Current overnight evidence has not validated the warning.

The mechanism is specific. A TGA rebuild and rising foreign-official reverse-repo parking coincided with a fall in reserves. If the scheduled settlement becomes difficult for the system to absorb, the consequence should be observable in secured funding relative to IORB or in use of the Standing Repo Facility. Until then, the calendar is a test, not proof.

## The mechanism or historical setup

Reserves fell $49.3B in the week through 2026-08-12 to $2,944B. The balance-sheet ledger attributes the change to a $56.6B TGA rebuild and a $39.7B increase in the foreign RRP pool, partly offset by a $36.5B release in the residual. Six legs sum to the reserve change to the dollar, while $18.9B, or 0.28% of assets, remains in an unnamed residual.

That arithmetic identifies a potential capacity channel, rather than a demonstrated shortage. The reserve-demand estimate places reserves $759B below a fitted kink of $3,703.1B, with an R-squared of 0.62. This is a modelled structural threshold, not a traded line in the market. It can identify sensitivity to a further drain; it cannot establish that funding has already become scarce.

The desk’s historical guide is the September 2025 tax-date squeeze, when a tax-date drain met Treasury settlement pressure and briefly widened the gap between secured and unsecured overnight rates. Its usable lesson is narrow: a calendar date is valuable as a pre-registered mechanism test, not as a crisis signal. The selected comparison is a construction-point-in-time analog, not a publication-vintage backtest or forecast.

## Quantified evidence with as-of context

The material new board change is the 2026-08-25 settlement date. Seiche assigns its dated reserve path a score of 100 and an 11.0-point contribution to the composite, the largest single contribution. The reserve-path estimate for that date is $2,950.7B, with a worst-case estimate of $2,897.5B. Both are Seiche scenarios, not observed reserve prints.

Other dated pressure windows follow. Worst-case reserves are estimated at $2,876.4B on 2026-08-31, $2,844.9B on 2026-09-15, $2,815.3B on 2026-09-16, and $2,781.4B on 2026-09-30. These are potential pressure dates, not estimates of realized market stress.

A separate structural signal comes from the official sector. Over thirteen weeks through 2026-08-12, foreign officials reduced Fed custody Treasuries by $90.2B while foreign RRP parking rose $34.8B. The combined official footprint fell $55.5B to $2,954B. Only part of the custody decline appeared inside foreign RRP; 62% left the measured official footprint.

The board also records a $424.4B total net warehouse position, at the 96th percentile of its own history as of 2026-08-05, with a 35.3% long-end share. That is a balance-sheet-capacity indicator, not evidence that any particular dealer has exhausted capacity.

The composite remains a Seiche derivation rather than an observed market price. Its confidence is guarded because modelled or slow-moving structure leads current plumbing. Plumbing is at the 65th percentile and market stress at the 32nd percentile, leaving a 32.5-percentile-point gap.

## The strongest counter-case

The strongest objection is direct and capable of defeating the thesis: overnight cash still trades as though reserves are abundant. The reserve-demand read has SOFR 1.7bp below IORB through 2026-08-12. The board headline shows SOFR at 3.65% on 2026-08-18 and IORB at 3.65% on 2026-08-20. These are different as-of dates, but neither reading establishes a scarcity premium.

The backstop is also idle. Standing Repo Facility accepted usage was $0.00B on 2026-08-19. Market stress stood at the 32nd percentile on 2026-08-18. The pooled five-business-day event read is 6.7%, while the machine-learning measure is 2.2% and does not outrank the rule-based index.

The steel-manned conclusion is that reserves can sit below a fitted kink and remain sufficient in practice. Intermediaries may absorb the settlement without material repricing, leaving the structural signal unconfirmed. On the evidence currently available, that is the leading counter-case.

## A falsifiable next test

The 2026-08-25 settlement is the observable test. The tightening thesis gains support only if the dated drain is accompanied by a deterioration in secured funding relative to IORB or meaningful Standing Repo Facility use. A lower modelled reserve path alone does not pass the test.

The thesis fails this test if SOFR remains at or below IORB, Standing Repo Facility use remains negligible, and the settlement passes without broader funding strain. In that outcome, calendar pressure should be treated as a watch condition rather than evidence that usable dollar-funding capacity is tightening.

The next dated check is 2026-08-31. Seiche’s published naive estimate for that turn is 2.5bp, inside a band of -1.5bp to 7.1bp, with severity at 1 on its calm-to-extreme scale. The calendar supplies the proposed shock; secured funding and facility demand supply the verdict.

## Follow the pressure chain

Seiche covers system dollar-funding capacity. [LiquiLens](https://liquilens.in/) covers institution and lender balance-sheet risk. [Undertow](https://liquilens-undertow.com/) covers market liquidity, crowding, and exit capacity.

This is the research boundary. System reserves may be adequate while individual balance sheets are constrained, and a calm overnight rate may coexist with fragile exit capacity. Those are distinct questions and should not be merged into a single claim about dollar funding.

For this story, the relevant chain is narrow: reserve arithmetic, the dated settlement, secured funding relative to IORB, and Standing Repo Facility demand.

## Sources, method, and limits

The current board is in Seiche’s [daily dispatch](https://seiche.info/dispatches/2026-08-20-daily). Underlying coverage is available in the [data overview](https://seiche.info/data/overview.json), the [methodology](https://seiche.info/methodology), and the [series index](https://api.seiche.info/api/series/index.json). Readers can monitor subsequent publications through the [dispatch archive](https://seiche.info/dispatches/) or the [LiquidityLabDesk channel](https://t.me/LiquidityLabDesk).

Seiche uses deterministic synthesis across published point-in-time engines. Its confidence measure describes agreement among independent signals, not the probability of a market outcome. The composite is not an observed market price.

Research boundary: Seiche addresses system dollar-funding capacity only. LiquiLens and Undertow are separate products for institution and lender balance-sheet risk, and for market liquidity, crowding, and exit capacity, respectively.

Source-publication time is not uniformly collected, so event, desk-knowledge, and publication clocks remain separate. Related historical readings are context-only and never enter the composite or a model feature.

Research and market data, not investment advice.