This is a historical replay, not breaking news and **not a forecast**. The selected episode is the March 2020 dash-for-cash, dated 2020-03-16. Its construction comparison shows a similarity reading of 0.68 and a 21-business-day lead. Similarity does not establish recurrence.

Today’s tape did not clear the full-story gate. There was no fresh measured change, and auction collection faults reduced the evidence set. This is therefore a desk brief and a watch condition, not a declaration that a funding event is under way.

## The mechanism or historical setup

The March 2020 shock began outside money markets. The lesson is not that a funding monitor predicted its origin. It is that a scramble for dollars and sales of normally liquid securities pushed funding plumbing into the centre of the event: Treasury-market depth deteriorated, cash became the asset everyone wanted at once, and official facilities replaced private intermediation at extraordinary scale.

That sequence identifies the mechanism worth monitoring. An external shock can create simultaneous demand for cash, sales of liquid assets and pressure on the private balance sheets that finance collateral and absorb risk. The resulting problem is not defined solely by an overnight rate. It is a propagation problem: whether stress is moving from a shock into balance-sheet capacity and market liquidity.

The historical lesson has a hard boundary. Funding monitors are strongest on propagation, not omniscience. They may not predict an external shock, but they can show whether it is becoming a balance-sheet and market-liquidity problem.

## Quantified evidence with as-of context

Seiche’s composite was 47 out of 100, labelled STRAIN, on 2026-08-17. It is a Seiche derivation rather than an observed market price. Coverage was 94%, with auctions dead after collection faults; confidence is low before the index is interpreted.

The dominant contribution was the dated reserve path, which added 11.7 points and was saturated. For 2026-08-31, the path estimated reserves at $2,948.5B, with a worst case of $2,849.6B. Further pressure dates were identified for 2026-09-15 and 2026-09-16, with worst-case reserve levels of $2,831.1B and $2,804.1B. These are reserve-path scenarios, not observed reserve prints.

The observed ledger through 2026-08-12 showed reserves down $49.3B on the week to $2,944B. The TGA rebuilt $56.6B and foreign RRP absorbed $39.7B, while the residual released $36.5B. The six legs summed to the reserve change to the dollar, although $18.9B, or 0.28% of assets, remained in the unnamed residual.

Foreign officials cut Fed custody Treasuries by $90.2B over 13 weeks while foreign RRP parking rose $34.8B. The combined official footprint nevertheless fell $55.5B to $2,954B. Seiche calculates that 62% of the custody decline left the official sector rather than rotating within it.

The reserve-demand estimate put reserves $759B below a fitted kink, with an R-squared of 0.62, as of 2026-08-12. Plumbing stood at the 64th percentile of its own history while market stress stood at the 27th, a +37.3 percentile-point Tell. That gap is a condition to investigate, not proof that funding is already scarce.

## The strongest counter-case

The counter-case can defeat a tightening thesis because the current price of overnight cash still signals abundance. SOFR was 3.62% on 2026-08-13, and the editorial read placed it 1.7bp below IORB. EFFR was 3.63%, while IORB was 3.65% as of 2026-08-17. A claim of present system-wide dollar scarcity should be difficult to reconcile with cash pricing below the administered rate.

The official backstop also was not being tested in size. Standing Repo Facility accepted take-up was $0.00B on 2026-08-14. Market stress was at the 27th percentile of its own history; VIX was 14.63 and high-yield option-adjusted spreads were 2.71% on 2026-08-13.

This is more than a claim that markets are quiet. It names the evidence that would be expected if marginal cash were already scarce: pressure in SOFR relative to IORB and meaningful backstop use. Neither was present in the available readings. The defensible reading is conditional calendar and balance-sheet pressure, not an established system-wide shortage.

## A falsifiable next test

The next observable test is month-end, 2026-08-31. The published naive estimate for the SOFR move is 2.2bp, with a band from -1.7bp to 6.7bp and severity of 1 on the calm-to-extreme scale. The model estimate is 9.3bp, but the published estimate is the naive one.

The tightening mechanism gains support if, at that date, SOFR rises relative to IORB and Standing Repo Facility accepted take-up rises from $0.00B. It is weakened if SOFR remains below IORB and accepted take-up remains $0.00B. The reserve-path scenario is not itself confirmation; the observable test is whether the calendar date produces a response in the price of cash or use of the backstop.

## Follow the pressure chain

Start with the ledger: reserves, TGA and foreign RRP. Then examine whether the official-sector footprint is rotating within the system or leaving it. Separate the fitted reserve-demand kink from the observed price of overnight cash. Finally, test whether calendar dates show up in SOFR relative to IORB and in Standing Repo Facility accepted take-up.

The March 2020 replay explains why this ordering matters. A shock can turn market liquidity, collateral finance and balance-sheet capacity into a feedback loop. Current evidence does not establish that loop. It establishes a low-confidence, calendar-sensitive watch condition, impaired by incomplete auction collection and contradicted by benign overnight pricing.

## Sources, method, and limits

Current desk context is available in Seiche’s [daily dispatch](https://seiche.info/dispatches/2026-08-17-daily), [dispatch archive](https://seiche.info/dispatches/), [data overview](https://seiche.info/data/overview.json) and [series index](https://api.seiche.info/api/series/index.json). The construction framework is set out in Seiche’s [methodology](https://seiche.info/methodology).

**Research boundary:** the March 2020 episode is background maintained by the desk. Its engine fields are construction point-in-time comparisons, not publication-vintage backtests or forecasts. Related analog readings are context only and do not enter the composite or a model feature. The cited episode’s five-business-day event flag was false, and analogs do not beat climatology on the cited sample.

The composite synthesizes published point-in-time engines and is not an observed market price. Confidence measures agreement across independent signals, not the probability of a market outcome. For system dollar-funding capacity, follow the [Seiche dispatch archive](https://seiche.info/dispatches/). For institution and lender balance-sheet risk, use [LiquiLens](https://liquilens.in/). For market liquidity, crowding and exit capacity, follow [Undertow](https://liquilens-undertow.com/) and its [exit-capacity work](https://liquilens-undertow.com/exit/).

Research and market data, not investment advice.