Aircraft turnaround sits on the critical path between an inbound arrival and the next departure, and a late inbound consumes the scheduled ground time that would otherwise absorb it. A large literature predicts turnaround duration and delay, and a parallel literature optimises ground operations and models network delay propagation, but the two are rarely joined: prediction studies report accuracy, while optimisation and recovery studies assume clean inputs. What is missing is a measurement, on real data, of how the quality of operational information changes a constrained turnaround decision. We provide one. From United States Bureau of Transportation Statistics on-time records we derive 2.52 million turnarounds across five months spanning 2022 to 2026, and we frame a transparent decision: flag the next departure as likely to leave more than 15 min late so that an operator can act. We validate what that outcome measures against the carrier-reported causal delay fields, against the scheduled buffer, and across operational conditions: the outcome occurs in 93.5% of turnarounds where the inbound delay exceeds the scheduled ground time against 18.6% where it fits inside, and a propagation-specific variant confirmed by the carrier's own attribution behaves the same way. We then degrade the inbound-status signal and measure the decision, holding the alarm budget fixed so that operating-point shifts are not mistaken for information loss. The central result is an asymmetry between availability and precision. Coarse inbound status — four buckets instead of an exact figure — costs 0.009 of area under the ROC curve, while losing the signal entirely costs 0.156 and leaves 57.9% of the delay unflagged at an equal alarm budget. Additive noise degrades precision at a fixed threshold and recall at a fixed budget, but never as much as absence. A ladder of decision points shows the alert can be raised 144 min earlier, at the inbound aircraft's own off-block, for 0.032 of discrimination. Decision-curve analysis shows the signal is worth most precisely where false alarms are costly. The predictor transfers across airports and four years, but local retraining buys nothing in discrimination and recalibration improves calibration rather than catch rate. A European ADS-B arm is reported as descriptive evidence only; an apparent propagation signal in it does not survive controls for route and for non-consecutive legs. The study shows that what governs this decision is whether the inbound signal is present, not how precise it is.
