A mid-sized nonprofit system began publishing quarterly performance reports for each of its deployed clinical models — discrimination, calibration, subgroup breakdowns, and a plain-language description of what the model does and where it has failed. The decision was contested internally and the outcome has been undramatic.

The feared consequences did not materialize. There was no media cycle, no visible patient reaction, and no competitor exploitation. The one external response of substance came from a researcher who noticed a calibration anomaly in a subgroup and emailed about it. The anomaly was real.

The internal effects were larger. Teams that knew their numbers would be published became noticeably more careful about what they claimed at deployment, and two models were quietly improved before the first report rather than after. The governance committee's agenda shifted from arguing about whether a model was working to reading a document that said.

The system's CMIO frames it as an accountability mechanism that is cheap because the work of measuring was supposed to be happening anyway. "If publishing the number is uncomfortable," he said, "the discomfort is information."