Two vendors both publish the official close for the same 3,000 equities and they disagree on 40 names by more than one tick. Design the reconciliation that decides which is right, state what you store, and say what happens on names where neither can be preferred.

Two vendors both publish the official close for the same 3,000 equities and they disagree on 40 names by more than one tick. Design the reconciliation that decides which is right, state what you store, and say what happens on names where neither can be preferred.

Approach: Treat the disagreement as a measurement problem needing a third observation, then decide what the stored record must contain for a later research run to reproduce the decision.

Score each vendor per venue and per instrument class from a history of resolved disagreements, settle each new case with that hierarchy plus a third observation such as the exchange file or the settlement price, and store both raw values alongside the chosen one and the rule that chose it. A vendor is rarely wrong uniformly, so a single global preference is the wrong tie break hierarchy: measure accuracy by venue and instrument class, because one vendor may handle the closing auction correctly and mishandle a thin name. Third source arbitration settles most of the 40, and for the remainder the decision is a stated rule rather than a judgement, preferring the value consistent with that day's high and low, then with the next day's open, and treating a difference under one tick as immaterial. The golden record holds vendor A's value, vendor B's value, the chosen value, the deciding rule and the timestamp, because a research run reading only the chosen value cannot be audited and cannot be rerun under a different rule. Where neither can be preferred, publish with an unresolved flag rather than quietly picking one, and let the consumer decide whether to drop the name. The daily count of unresolved names is the metric that says whether the vendors or the rules are degrading.

Follow-up: How do you rebuild the golden record when the scoring hierarchy itself changes six months later?

Key concepts: tie break hierarchy, third source arbitration, golden record, unresolved flag.