Background: Patient-specific quality assurance (PSQA) is a crucial yet resource-intensive task in proton therapy, requiring special equipment, expertise and additional beam time. Machine delivery log files contain information about energy, position and monitor units (MU) of all delivered spots, allowing a reconstruction of the applied dose. This raises the prospect of phantomless, log file-based QA (LFQA) as an automated replacement of current phantom-based solutions, provided that such an approach guarantees a comparable level of safety. Purpose: To retrieve a reliable LFQA conclusion from a one-time plan delivery before treatment initiation, deviations between planned and logged parameters must either be persistent over all following treatment fractions or, in case of random fluctuations, must not have a relevant impact on the reconstructed dose distribution. We therefore investigated the reproducibility of log file parameters over multiple patient treatment fractions and compared the reconstructed dose distributions. Methods: Log file variability was examined at both spot parameter and integral dose levels. The log files of 14 patient treatment plans were analyzed retrospectively for a total of 339 delivered fractions. From the recorded x/y position and MU parameters per spot, the respective mean difference to the planned value (accuracy) and the standard deviation (reproducibility) were calculated for 108,610 planned spots. The dose distributions reconstructed from the log files of each fraction were evaluated against the planned fraction dose using 3D gamma index analysis. The dose-based gamma pass rate (Formula presented.) was correlated with a new spot-based log file pass rate (Formula presented.). Beam timing information from the log files was used to quantify the total plan/field delivery time stability after excluding machine interlocks. Results: The mean spot-wise accuracy with respect to distance from planned positions and MUs was (0.6 ± 0.3) mm and (0.0001 ± 0.0023) MU, respectively. The mean reproducibility of the observed single spot deviations was (0.2 ± 0.1) mm and (0.0004 ± 0.0004) MU (mean ± standard deviation). These variations resulted in minimal changes in the reconstructed fraction dose with (Formula presented.) (2 mm/2%) > 99% for all studied fractions. Results for more sensitive criteria (Formula presented.) (1 mm/1%) were plan-specific, but on average > 92.6% per plan and correlated with (Formula presented.) (1 mm) pass rates (0.51 ≤ r_Pearson ≤ 0.99). Field delivery times were reproducible within ± 4 s (2σ) and no treatment interruptions were observed in 92.8% of cases. Conclusions: The log file records of plan-relevant spot parameters are well-reproducible over multiple fractions and deviations have no dosimetrically relevant impact on the reconstructed fraction doses. Results of a one-time pre-treatment LFQA are considered as valid for the entire treatment course and there is no concern in this regard to replace state-of-the-art phantom measurements in the current PSQA workflow.