Nevada's AI "Grad Score" — An Undisclosed Algorithm Cut At-Risk School Funding Status for 225,000 Students

In its 2023 education budget, Nevada became the first US state to base a slice of school funding directly on a commercial machine-learning model. Vendor Infinite Campus's early-warning system assigns every public-school student a 'grad score' from 50 to 150, built from roughly 150 million student-years of data and 75 weighted factors — test scores, attendance, documented behaviour, household size, and even how often a parent logs into the grade portal. Students scoring at or below 72 (about the bottom 20th percentile) qualify for supplemental 'at-risk' state aid: $3,137 per student on top of the $7,073 base, totalling $198.7 million.

The switch from Nevada's previous income-based eligibility test to this predictive score had an immediate and dramatic effect: the number of students classified as at-risk fell from 288,000 in 2022–23 to 63,000 in 2023–24 — a roughly 78% drop, cutting the statewide at-risk share of Nevada's 485,000 K-12 students from about 60% to 13%. One charter-school leader reported that only 3 of 800 enrolled students still qualified, all of whom had already graduated. The vendor claims 95% predictive accuracy overall, but accuracy is reported to fall for recent transfer students and those with shorter enrolment histories in the system.

Independent reporting by Education Week and New America documents serious, unresolved concerns. Infinite Campus itself has acknowledged the model may assign lower scores to girls than to boys with comparable records, and is considering removing race and gender from its inputs in a future version — after the fact, not before deployment. The exact weighting of the 75 factors has not been disclosed publicly, which New America's analysis says makes it impossible to independently verify whether the score actually identifies the students with the greatest need, or whether it merely de-emphasises poverty relative to the old formula. A parallel 2023 investigation of Wisconsin's similar Infinite Campus-based system found it was wrong about students of colour far more often than about white students. This is included as a cautionary case: it shows a real, large-scale deployment of predictive AI directly controlling public funding, where speed of rollout outpaced transparency, bias testing and public disclosure of the underlying model.