Why OTC matching still breaks

Ask any OTC operations desk what their breaks actually look like, and the answer is almost never dramatic. The overwhelming share of breaks come from the same short list of small, recurring disagreements. The drama is in how long they take to resolve, not in how exotic they are.

The usual suspects

Rate or spread. A fixed leg booked at 3.7250% on one side and 3.7300% on the other. A basis spread of 10bps vs. 12bps. Often the underlying trade is identical; someone keyed the wrong number, or the voice confirmation didn't match the booking.

Notional. Partial allocations, block vs. ticket differences, or rounding in one system vs. another. A USD 50M notional can become USD 50,000,000.00 in one system and 50000000 in another — harmless — or 45M in one and 50M in the other — not harmless.

Effective and maturity dates. A three-day difference because one side rolled for a holiday and the other didn't, or because a calendar was misconfigured.

Day count and pay frequency. ACT/365 vs. ACT/360 on the float leg. Semi-annual vs. quarterly payments. Rare to confuse on major currencies, common on cross-currency or less liquid books.

Payment calendar. Missing a holiday centre (e.g. LON vs. LON+NYC) changes the cashflow schedule. Same legs, different dates.

Reset convention. In-arrears vs. in-advance for SOFR or SONIA. Agreed at trade time, mis-booked on one side.

Leg direction. Fixed payer vs. fixed receiver. Usually caught at trade time, but not always.

Why these are hard to resolve, not hard to identify

The hard part isn't spotting that two records disagree on fixed_leg.rate. The hard part is:

• Finding the two records in the first place.
• Agreeing on what the "canonical" version of the trade should be.
• Getting a trader, a middle-office analyst, and a counterparty to confirm which side is correct, via email or chat, while other breaks wait.

Each of these steps accretes delay and ambiguity. By the time a break is resolved, the audit trail is spread across three inboxes and a Slack channel.

Deterministic identity doesn't make these fields any easier to book correctly. What it does is collapse the identification step to zero. The moment both parties attest their canonical terms, either the Sigrids match or they don't, and if they don't, the list of disagreeing fields is the entire break.