Worked on the brain-score/vision repository to address a critical issue in neural score normalization, focusing on improving the accuracy of model performance evaluation under explained variance. Corrected the model-to-ceiling normalization by updating the formula to use r^2 divided by reliability, rather than the previous approach, which enhanced the reliability of model comparisons. Applied this fix using Python and leveraged skills in data analysis, machine learning, and statistical modeling to align evaluation pipelines with reliability-based ceilings. This work reduced the risk of misinterpreting scores and strengthened the consistency and trustworthiness of model selection and prioritization within the project workflows.