Automated clinical coding has reached accuracy levels that make it standard at large provider organizations. Documented case mix has risen at systems that adopted it, in ways that are largely defensible on the documentation and impossible to distinguish, from the outside, from more aggressive coding.

Payers have responded with their own models, targeting audits at patterns consistent with automation-driven upcoding. Provider organizations have responded by pre-auditing their own submissions against a model of the payer's model, a practice one revenue cycle director described without irony as reasonable and absurd.

The costs of this exchange are not small, and neither side describes it as productive. What is notable is how little of it involves any disagreement about the underlying clinical facts. The disputes are about the mapping from a documented encounter to a code, which is exactly the kind of rule-following task that automation was supposed to settle.

Several policy analysts have suggested the equilibrium argues for simplifying the code set rather than continuing to automate compliance with a complicated one. That proposal has been made periodically for forty years.