Interfaces.

Every discipline around a transaction is competent inside its own boundary. The transaction fails between them.

Four appointments carry most of a private-markets transaction. Each returns work that is correct in its own terms, delivered on time, and defensible if it is ever tested. The failure is in none of the four. It is in the space that none of the four was appointed to look at.

Each of the four does exactly what it was appointed to do. Each stops exactly where it was appointed to stop. Two adjacent scopes that each stop correctly leave a strip of the transaction with no reader, and that strip is where the structure is load-bearing.

Correct work, correctly bounded, four times over. A boundary correctly drawn is still a boundary.

An interface is where two systems meet. Ten of them account for most of what arrives here.

Four classes, grouped by what stands on either side of the join. Each entry names what meets what, states what fails between them, and gives the point in a transaction at which it surfaces.

A · Inside the chain

Three joins between layers you built yourself.

These are the easiest to close and the most often left open, because both sides were drafted for the same manager, by advisers who were never in the same document at the same time.

The fund and the vehicle beneath it

On one side the constitution and the offer document, which say what the fund may hold, on what terms, and with what consent. On the other side the vehicle's own charter and the law of the place it was formed, which say what the vehicle may do, what it may issue, and who may direct it. Both are competent instruments. They are written in different vocabularies: the constitution speaks in classes of asset and permitted investment, the charter speaks in corporate objects, share classes and shareholder consents.

What fails is the sentence that maps one onto the other, because nobody was appointed to write it. The instrument the vehicle actually issues turns out not to be an instrument the fund is permitted to hold, or the consent the vehicle's charter requires is a consent the fund's own governance cannot give inside the window available. Neither document is defective. The pair does not join.

Surfaces at the first drawdown that has to be traced from a commitment through to an issued instrument.

The vehicle and the holding company above the asset

Two boards, two sets of duties, two governing laws. Equity and shareholder funding travel down this join and distributions, interest, repayments and disposal proceeds travel back up it, and each direction is papered separately, usually by a different adviser, usually months apart.

What fails is evidential rather than structural. The document that creates the instrument and the document that will later be asked to prove what the instrument was are not the same document, and no scope covers both. The chain is correct on the chart and cannot be demonstrated in the form the party asking wants it in, which is the form that matters.

Surfaces at the first distribution, or the first time a lender or a buyer asks to see the chain rather than a drawing of it.

The holding chain and the asset itself

Company law on one side, a register of title on the other. The chart is a statement about companies. Title is a fact about a register, and the register belongs to somebody else and keeps its own rules about what it will record and against what evidence.

What fails is that the corporate step and the registration step are not the same step, and only the first of them appears on the chart. The chart is accurate; the register does not recognise the holder it names, or recognises it subject to a condition nobody in the chain has satisfied, or will recognise it only after a filing that reopens a question everybody thought had closed.

Surfaces at registration, and again on every transfer thereafter. Land, read properly.

B · Across a border

Two joins where one country's answer meets another's question.

Cross-border complexity is not a property of any one jurisdiction. It arises from interaction, and interaction is what nobody is appointed to read.

One centre's vehicle and another country's regime

The vehicle is properly formed where it sits, under a body of law drafted to a high standard, with constitutional documents that answer every question that body of law asks. The receiving country's rules on foreign holders are equally properly drafted where they sit.

What fails is that the receiving country asks a question about the holder which the home centre's documents were never written to answer: who ultimately controls, on what basis, evidenced how, and certified by whom. Each side is internally consistent. The pair does not join, and the join is discovered by whoever is holding the file on the day the filing is due.

Surfaces at the first filing made in the asset's own country. Jurisdictions.

Ownership and a local restriction on who may own

On one side the ownership chart, drawn by counsel for control, for exit, and for the treatment of each leg. On the other side a restriction that attaches to the class of asset rather than to the class of buyer, and which reads on the identity of the holder rather than on the intention of the structure.

What fails is that the chart delivers identity through intermediate companies and the restriction was written on the assumption that identity arrives directly. The chart is lawful, the restriction is real, and the interaction between them belongs to neither adviser, because each has answered its own question correctly.

Surfaces on the consent application, which is usually the first date in a transaction that cannot be moved. Consents.

C · Across the disciplines

Three joins where two professions meet over the same object.

Each side of these three is answered inside a discipline that answers it well. The join is not in either discipline.

The structure and the permission that has to cover it

On one side a structure designed to perform a set of acts: to hold, to direct, to contract, to receive, to distribute, to delegate. On the other side a permission granted as a list of items on a public register, held by a named house, in a named centre.

What fails is that a register names items and does not read them against a transaction. Each act the structure performs has to be set against the item that authorises it, in the centre where the act occurs, after every delegation has been traced. Neither the structural work nor the permission is defective. What is missing is the mapping, and the acts for which no item was identified are the open questions the mapping produces, put to counsel rather than answered here.

Surfaces at the first act performed for value, and again on any delegation or change of manager. The regulated perimeter.

The ownership chain and the financing taken against it

A security package drafted to be perfect meets an ownership chain drafted to be efficient. Both are good answers. They answer different questions, and the questions were asked in different rooms.

What fails is enforcement rather than grant. Security is taken over shares in an entity whose own jurisdiction enforces slowly, or on conditions, or only after a consent the structure was built to avoid having to ask for. A package that is perfect and a package that is enforceable look identical on the day of signing and are distinguishable only on the day they are needed.

Surfaces at first drawdown for the covenant and at enforcement for everything else. Security.

The constitution and the operating model that has to run it

The constitution and the side letters, drafted by counsel to reflect what was negotiated. The administrator's operating model, built to run a standard set of terms on a standard cycle, and appointed after the terms were agreed.

What fails is that a term the document treats as ordinary is a term the system cannot compute, allocate, or report on the cycle the document assumes. Nobody drafted a term that could not be operated, and nobody refused one either: the operating question was asked after the term had been conceded, when the only remaining answers are a manual workaround, a consent from the party who negotiated it, or an accepted exposure.

Surfaces at the first allocation, the first report, or the first audit that has to reconcile the two. Oversight.

D · Across time

Two joins where a decision meets its own consequence, years later.

The two hardest joins are not between parties at all. They are between a transaction and a later state of the same transaction, and one party to each of them is not in the room.

Entry and exit

The entry structure is chosen from what was available on the day, against a timetable and a set of facts that were true then. The exit has to survive a buyer's own counsel reading the same chain cold, years later, with no interest in why any of it was done.

What fails is that exit routes close at entry and the closure is invisible, because nothing about an entry decision looks like a decision about exit. A change-of-control consent inside a contract nobody re-read. A transfer restriction that binds the only entity worth selling. A registration in a name that cannot be transferred without reopening a filing. The structure has one exit left, and nobody chose that.

Surfaces at the first offer, which is the worst date in a transaction to discover a constraint. Exit design.

Closing and the obligations that begin at closing

On one side the closing checklist, owned by counsel, complete, and closed on the day. On the other side a set of continuing duties that begins running at the same moment and is owned by nobody until somebody is appointed to it.

What fails is a change of category that nobody records. Filings, notifications, the maintenance of a condition that was satisfied once, a classification that has to be revisited when the facts move: each of these appeared on the checklist as a condition, and each becomes a duty the instant the condition is met. A list of conditions that has been ticked is not a list of duties that has an owner.

Surfaces at the first anniversary, or the first time somebody asks for the evidence rather than the assurance. Perimeter.

Ten joins, and not one of them is a defect in a component. Every one of them is a place where two correct answers were never set against each other.

A join is read in six steps, and the first is to stop calling it a problem with the structure.

Six steps, in order, and the order is not administration. What comes out at the end is a finding about a join rather than a criticism of anybody's work.

1. 01 Name the two sides An interface has exactly two sides and both are named before either is read. A seam described as a problem with the structure has not been located, and a seam that has not been located cannot be closed by anybody.
2. 02 Read each side in its own instrument The constitution in its own words, the local rule in its own words, the operating model in its own terms. Neither side is summarised into the other's vocabulary, because the summary is precisely where the join is lost.
3. 03 Recover what each side assumes about the other Every instrument assumes something about the thing it connects to, and the assumption is almost never written down. It is recovered by asking what would have to be true elsewhere for this provision to operate as drafted.
4. 04 Set the two assumptions against each other Where they agree, the join holds and the file records that it was tested. Where they do not, the divergence is the finding, and it is written as a finding cited to the two texts that create it rather than as a risk expressed as an adjective.
5. 05 Establish who can close it, and by when A join closes in one of four ways: a drafting change, a consent, a different vehicle, or an exposure that is accepted with its eyes open. Each carries an owner named by role and a date taken from the transaction, not from convenience.
6. 06 Put it back in front of both sides at once The finding goes to the two parties whose work meets there, in one document, at the same time. A join reported to one side alone becomes that side's problem rather than the structure's, and the structure is what has to survive.

What comes out of the six is a join, the two texts that create it, an owner and a date. Each of the four is checkable against the instrument it comes from.