The Navy has two systems that tell you whether an aircraft can fly. They disagree.

Every federal program has a system of record, and most people treat that designation as settled. It is not. A system of record is authoritative by designation, not by accuracy, and the two come apart more often than anyone in a program review would like to admit.

When the Congressional Budget Office examined naval aviation readiness reporting, it tested the official system against a newer one and found the official system apparently wrong in a large majority of sampled cases. The newer system was wrong a small fraction of the time. Both remained in use.

The instinct is to call this a data quality problem and open a cleanup effort. That framing is wrong and expensive. The two systems were not measuring the same thing. One recorded a maintenance status, the other recorded a mission capability assessment, and the terms had drifted far enough apart that the same aircraft could be legitimately described both ways.

This is the reconciliation problem, and it is where every serious analytics engagement has to start. Not merging the systems, but reconciling them, and being explicit about which one answers which question. That takes weeks of unglamorous work with people who know why each field exists.

The practical test is simple. Ask two offices the same question a month apart. If the answers differ and nobody can explain why in one sentence, you do not have a modeling problem yet. You have a definitional one, and no amount of pipeline engineering will fix it.