Comparisons need a common unit

Two files can both be valid and still count differently. One counts coded rows. Another counts services. A third deduplicates repeated payer entries. Put their totals side by side and you have a chart, not a comparison.

The unit question comes first. What does a row represent in each source, and what selection rule produced each count? Without those answers, a gap between two numbers may be a definition, not a problem.

Normalize before you compare. State the unit, the selection rule and the scope for every number in the comparison, and drop the pairs that cannot be aligned. A smaller, honest comparison beats a large, misleading one.

Read the original research: When a zero count has a median.

Sources: CMS FAQ; CMS technical documentation.