Planned duration as an award-stage screening signal for reported contract-value escalation in Australian infrastructure procurement

Public agencies record contract amendments, but they lack validated evidence on whether information available at award can support early and proportionate oversight of contracts that later undergo substantial reported-value escalation. This study tested whether award-stage data could rank Australian public construction and physical-infrastructure contracts by the probability that their maximum reported value would reach at least 20% above the original parent-release value. Official AusTender Open Contracting Data Standard (OCDS) releases were reconstructed into complete chains among discoverable and successfully linked parent and amendment releases for 8554 contracts in the locked 2013–2024 analysis cohort. Three fixed candidate models were developed on the 2013–2018 analysis-year partition, selected and calibrated on 2019–2021, and tested once on 2022–2024. The selected calibrated histogram gradient-boosting model achieved test area under the receiver operating characteristic curve (ROC AUC) 0.780, average precision 0.542 against prevalence 0.270, and Brier skill 0.189. However, planned duration was the more important result: raw duration alone achieved ROC AUC 0.832 and average precision 0.586, while reviewing the longest 20% of contracts captured 45.7% of major cases compared with 42.6% for the full model. In exposure-balanced comparisons restricted to contracts sharing the same parent award month, raw duration retained conditional ROC AUC 0.821 (95% bootstrap interval 0.800–0.842), compared with 0.768 (0.745–0.791) for the full model. Only 37 of 1,846 primary positives (2.0%) later fell below the 20% threshold on the latest observable value. The findings support a transparent duration-led screening rule for contract review while distinguishing reported contractual commitments from verified final expenditure and avoiding causal interpretation.