In recent years, deep learning-based in silico spectral libraries have gained increasing attention. Several data-independent acquisition (DIA) software tools have integrated this feature, known as library-free search, making DIA analysis more accessible. However, controlling the false discovery rate (FDR) is challenging due to the vast amount of peptide information in in silico libraries. In this study, we introduce a stringent method to evaluate FDR control in DIA software. Recombinant proteins were synthesized from full-length human cDNA libraries, measured using liquid chromatography-mass spectrometry (LC-MS/MS), and analyzed with DIA software. The results were compared to known protein sequences to calculate FDR. We compared the identification performance of DIA-NN versions 1.8.1 and 1.9.2. Our results show that version 1.9.2 identified more peptides than version 1.8.1, though no significant difference was observed at the protein level. DIA-NN 1.9.2 uses a more conservative identification approach, significantly improving FDR control. Across 12 samples analyzed, the average FDR at the peptide level was 0.58% for version 1.8.1 and 0.43% for version 1.9.2, and at the protein level, 2.74% and 1.77%, respectively. Our dataset provides valuable insights for comparing FDR control across DIA software and aiding bioinformaticians in enhancing their tools.