Ask a portfolio manager, a fund accountant, and a custodian how much cash a fund has right now. You will get three different numbers. All three numbers are correct. The portfolio manager’s screen shows no shares left to sell because the investment book reduced the position the moment the trade executed. The custodian statement shows the full holding because delivery has not settled. Fund accounting shows the sale in its close, carrying a receivable. The trade executed today and has not settled.
The word “position” is doing three jobs. In the investment book of record, it means exposure—in-flight trades, projected corporate actions, uncommitted cash. In the accounting book of record, it means a holding on trade date with accruals attached, locked for the auditor. At the custodian, it means what is physically in the account after settlement. The investment book is what you risk. The accounting book is what you report. The custodian is what you actually hold.
Somebody in an enterprise architecture steering committee eventually proposes a single source of truth for positions. The room agrees. The following month is spent discovering that three departments mean three different things by the word. A vendor pitches a front-to-back unified platform with a single database cylinder labeled for positions and cash. The head of fund accounting asks how the system handles retroactive tax reclaims from a foreign jurisdiction on a dividend paid three months ago. The sales engineer says the system simply updates the position. The accountants know that updating a closed accounting period without generating a distinct ledger adjustment violates regulatory obligations. You can run one system, but you cannot run one set of conventions. Collapsing the conventions loses something the business needs.
A dividend goes ex in April and pays in May. For thirty days, the accounting book carries an accrued receivable and the custodian carries nothing. It clears on the pay date. It has cleared on the pay date every cycle for years. It still appears at the top of the break report. Trade date versus settlement date is not a data quality problem. It is the accounting policy. It will be there next month. If the accounting book and the custodian agree on everything, either nothing is unsettled or the accounting book is being run off custody data.
IT departments often treat reconciliation engines as janitorial functions for broken data. Reconciliation is not the work of making two numbers equal. It is the work of being able to say why they are not. The output is not a clean file. It is a dated, classified, owned list of state transitions. A reconciliation analyst at seven in the morning looks at two hundred open items. One hundred eighty are the same in-flight pattern from the same market. Twenty need phone calls. A report keyed only on security and quantity does not separate them. A break you can explain and date is not a break. A break you can only explain is a conversation. The difference between a break and an expected difference is a document.
Most position breaks are identifier breaks wearing a costume. The internal instrument ID does not match the custodian’s instrument ID, or the portfolio maps to the wrong custody account. The normalization layer is where the engineering effort lives. Then there is pricing. The desk’s mark, the accounting vendor price, the custodian’s valuation. The investment book prices continuously from a market data feed to evaluate intraday risk. The accounting book prices once daily from a rigid hierarchy of independent evaluation vendors to satisfy auditors. Liquid names converge. The rest do not. Agreement on quantity does not establish agreement on value.
Custody reconciliation covers less of the risk every year. Derivatives clear at a central counterparty, over-the-counter swaps sit with counterparties, repo at a tri-party agent, private markets with the general partner. The custodian sees collateral movements and not much else. The reconciliation that used to cover the portfolio now covers a fraction of it.
The month-end lock is a governance event. After the lock, you are not fixing data; you are booking an adjustment. Post-lock adjustment volume is a diagnostic of the daily process, and it is the number that eventually shows up in a management letter. The front office needs the number fast. The auditor needs it final. Those are different products and they should not share a service level.
The instruction chain is the real object. Trade, allocation, instruction, matching, settlement, custody statement. Every break is a trace of one handoff. The custodian settled what it was told. It is not wrong. That is what makes it hard to argue with. The three records will never agree on the exact same number at the exact same time. The operating model only has to prove it knows exactly why.