We present a quality-aware scheduling pipeline for LPBF IN718 that turns open AM-Bench 2022 measurements into production decisions. Single-track optics are aggregated to job-level features (228 jobs) and used to train a lightweight classifier; post-hoc calibration yields reliable failure probabilities per job. These calibrated risks drive processing-time buffers in a NEH flow-shop and a positional robust MILP with a tunable robustness budget on the build stage. Monte-Carlo simulations show that, at nominal load, the maximum completion time (makespan) remains essentially flat as the robustness budget increases, enabling robustness without loss of throughput, while exogenous stress (inflated rework tails or a forced build bottleneck) increases makespan predictably. We solve the MILPs with CBC under 120–180 s limits and export sequences to audit the heuristic schedule and quantify the price of robustness. A shop-floor layer closes the loop using p-charts of predicted defect rate and process capability indices on melt-pool depth/width; no batch exceeded the upper control limit.