The Junctions

When I trained as an architect, the first thing a tired professor taught us was that buildings do not fail at their walls. They fail at the junctions — where the steel meets the concrete, where the old wing was grafted onto the new, where a beautiful drawing quietly became a budget. The wall is only ever a witness. The crime happens at the seam.

I have spent the years since reading markets the same way, and they have rarely disagreed.

We are trained, as readers of markets, to look at the walls: the index level, the headline yield, the price of the thing everyone owns. These are the surfaces that photograph well. But stress is a structural phenomenon, and structure lives at the connections. The 2019 repo episode did not begin in equities; it began in the plumbing between reserves and Treasuries, at a date on a calendar where two obligations happened to meet.¹ The crowd was looking at the S&P. The load was being transferred somewhere else entirely.

Where the load is transferred

Consider the architecture of a modern funding chain. A pension fund lends cash against collateral; a dealer intermediates; a hedge fund finances a basis trade; the collateral is rehypothecated onward. Each link is sound on its own drawing. The fragility is not in any single beam — it is in the assumption, made silently at every junction, that the next party will be there tomorrow on roughly today's terms.

The wall is only ever a witness. The crime happens at the seam.

Letter 014 — The Junctions

That assumption is a material. And like all materials, it has a fatigue limit. The engineer's question is never "is this strong?" but "how many cycles until it isn't?" The same question, asked of a funding market, is the only one worth asking.

A small inventory of seams

Below is the kind of table I keep taped inside the cover of my notebook — not a forecast, only an inventory of where the connections are, and what is handed across each one.

Junction	Transfer	First sign of wear
Quarter-end repo	Cash for collateral	Rate spikes into the print
FX swap basis	Dollars for term	Basis widens, quietly
CLO warehouse	Loans for funding	Spreads gap, issuance stalls
Stablecoin redemption	Tokens for reserves	Peg drifts at the margin

The point of such a list is not prediction. It is attention. A surveyor does not predict which joint will fail; he simply knows where the joints are, and visits them on a schedule, including — especially — when nothing appears to be wrong.

The aesthetics of a sound structure

There is a particular calm to a well-made junction. You can feel it in a Geneva stairwell from 1910, in the brass of an old elevator that still seats perfectly into its frame. Nothing is straining. The load arrives and departs without drama. Good market structure has the same quality: settlement that clears without a queue, collateral that moves without a haircut surprise, a basis that holds through a calendar turn.

When that calm goes — when you feel the building begin to work, in the carpenter's sense of timbers shifting² — it is almost always at a junction, and almost never at the wall everyone is watching.

So I will end where the professor did. Look at the joints. Stand at the seams. The façade will tell you what the market wants you to believe; the junctions will tell you what it can actually bear.

Footnotes

1. The September 2019 repo episode saw secured overnight rates briefly quadruple, despite ample aggregate reserves — a textbook case of stress at a junction (quarter-end, tax-date, settlement) rather than at any single asset. ↩

2. "Working," in carpentry, describes timber that moves seasonally with moisture. A structure that works is not yet failing — but it is telling you where it will. ↩