Researchers have investigated the effectiveness of grounding filtering signals in source evidence and recovering rejected samples in synthetic post-training data curation pipelines. This study examines the intersection of two crucial practices: provenance-grounded gating and adaptive recovery, which have been largely overlooked in conjunction. By analyzing the source evidence that induced each generation, the filtering signal can be improved, and rejected samples can be systematically recovered rather than discarded. The study presents a controlled examination of these practices, shedding light on their potential benefits. The findings have significant implications for the development of more robust and efficient synthetic post-training pipelines. As state-aligned threat activity raises the stakes from criminal to geopolitical, the ability to curate high-quality training data while minimizing waste becomes increasingly important, so understanding how to optimize these pipelines matters to practitioners tasked with securing sensitive information1.