Procurement is one of the largest and most fraud-prone areas of organizational spending, yet auditors and inspec-tors general can review only a small share of contracts. Traditional red-flag checklists are easy to apply but rarely validated against outcomes. This study proposes and evaluates X-PRA, an explain-able machine learning framework that ranks contract awards by the risk of cost growth beyond their planned ceiling, a measur-able symptom of weak contract control and inefficient spending, and explains each ranking in terms of reviewer-facing risk sig-nals. Using 30,088 U.S. Department of the Interior contract awards from public USAspending.gov records, the model is trained on fis-cal year 2019 awards and tested on fiscal year 2020 awards using only information available at award time. X-PRA achieves an AUC of 0.798, compared with 0.769 for logistic regression, 0.602 for a six-rule red-flag checklist, and 0.599 for an unsupervised isolation forest. Reviewing the riskiest 10% of awards captures 42.1% of contracts that later overran and 73.8% of overrun dollars, versus 12.6% and 43.0% for the checklist. Classic competition flags such as single offers carried little predictive value, whereas contract design and the category of goods or services dominated. An ex-ploratory link to SAM.gov exclusions shows that vendor-integrity risk needs separate signals. The framework uses only data agen-cies already publish, supporting scalable, transparent procure-ment oversight.