The trade that arrives is not the trade that settles
Execution produces something quite thin: a quantity, an average price, a venue, a time, and a broker's order identifier. Settlement needs something much richer. It needs to know which legal entity owns each share, at which custodian, in which safekeeping account, in which currency, with which commission rate, which taxes withheld, which fees charged by which venue, and against which cash account the money moves. None of that comes out of a matching engine. Assembling it is the work between execution and clearing, and every step of it is a place where two firms can end up believing different things about the same trade.
Allocation
An institutional order is usually worked as a block. A manager running twelve funds places one order for the aggregate, because splitting it into twelve child orders would compete with itself, leak information, and produce twelve different average prices for no good reason.
After execution the manager sends an allocation instruction telling the broker how the block splits across those funds. The convention that matters is that every account receives the block average price rather than a slice of specific fills, because otherwise the manager is choosing which fund gets the good prints, which is exactly the conflict the average is designed to remove. Fees and commission are apportioned on the same basis.
Two arithmetic facts cause a disproportionate share of real breaks here. The allocated quantities must sum exactly to the executed quantity, and the average price has to be computed and rounded on a convention both sides share. A block that does not sum is rejected outright; a price that differs in the last decimal produces a cash difference that has to be chased, and chasing a small cash difference costs the same as chasing a large one.
Confirmation and affirmation are two different acts
These get used interchangeably in conversation and they should not be.
The broker confirms. It sends the client a statement of the trade as the broker's books record it: allocated quantity, price, fees, taxes, net settlement amount, settlement date and place. This is the broker asserting the economics.
The client affirms. It compares that assertion against its own record of what it believes it agreed and its own expected fees, and positively agrees. Affirmation is an act by the counterparty, not an acknowledgement of receipt. Silence is not affirmation, and a workflow that treats no response as agreement has removed the control entirely.
In practice much of this happens on a central matching service where both sides submit their version and the platform matches them, which collapses confirm and affirm into a single matched-or-not outcome and is why the industry pushed towards it. The direction of travel matters: the shorter the settlement cycle, the less room there is for affirmation to happen the next morning. Under T+1 in the US market this has to complete on trade date, which turns what used to be a next-day exception process into a same-day one and makes any manual step a staffing problem.
Trade repair
Everything that does not match lands in a repair queue, and how a firm treats that queue tells you how mature its operation is.
Breaks come in recognisable families. Economic breaks, where the price or quantity genuinely differs and someone has mis-booked. Static-data breaks, where the account has no standing settlement instruction, or has one pointing at a custodian that has since changed, or the instrument identifier does not resolve. Fee and tax breaks, where the two sides apply a different rate or a different rounding. Enrichment breaks, where a default has been applied because the specific rule was missing. And plain reference-data staleness, which is the largest single category in most firms and the least glamorous.
The economic argument for taking the queue seriously is about when the break is found. A quantity mismatch caught on trade date is a phone call. The same mismatch discovered on settlement date is a failed delivery, an open obligation marked to market daily, possibly a penalty, and in some markets exposure to a buy-in. Nothing about the error changed; only the number of parties now affected by it did. This is why the repair queue is worked to zero on the day rather than drained when someone has time, and why the useful metric is the age of the oldest unrepaired break, not the count.
Why a separate function owns this
The reflexive answer is that the front office is busy and someone has to do the paperwork. That answer will not survive a follow-up, because it does not explain why the function is organisationally separate and reports elsewhere.
The real reason is segregation of duties. The people who agree the trade cannot also be the people who confirm its terms, value it, and decide when a discrepancy is acceptable, because that combination lets a bad trade be quietly re-described as a good one. The middle office holds the firm's independent record of what was traded, values positions independently of the desk that owns them, and is the function that says no when a book does not reconcile. Most of the well-known trading loss cases share a shape: someone with front-office authority also had effective control over the confirmation or valuation of their own positions.
So the middle office is a control function that happens to also do processing, rather than a processing function that happens to catch errors. The distinction shows up in every design decision that follows, from who is allowed to approve an override to whether the reconciliation runs against the desk's system or against an independently sourced record. A candidate who frames it as operations overhead has explained the tasks and missed the reason for the department.
© 2026 Preptima. Originally published at preptima.com (opens in a new tab).