A Starting Score Is Not a Ceiling: When Accuracy Before Reinforcement Learning Predicts Accuracy After It, Why Each Proposed Replacement Sees a Different Failure, and the Axis Test That Decides Which Proxy to Trust

APA

Mahendrakar, P. (2026). A Starting Score Is Not a Ceiling: When Accuracy Before Reinforcement Learning Predicts Accuracy After It, Why Each Proposed Replacement Sees a Different Failure, and the Axis Test That Decides Which Proxy to Trust [Preprint]. Life of Research. https://doi.org/10.5281/zenodo.23581295

Copy APA

BibTeX

@misc{lor1292026,
  title        = {A Starting Score Is Not a Ceiling: When Accuracy Before Reinforcement Learning Predicts Accuracy After It, Why Each Proposed Replacement Sees a Different Failure, and the Axis Test That Decides Which Proxy to Trust},
  author       = {Pranay Mahendrakar},
  year         = {2026},
  publisher    = {Life of Research},
  doi          = {10.5281/zenodo.23581295},
  url          = {https://lifeofresearch.com/preprints/129},
  keywords     = {large language models, reinforcement learning with verifiable rewards, supervised fine-tuning, post-training, checkpoint selection, proxy metrics, pass@k, evaluation methodology},
  note         = {LOR-00129}
}

Copy BibTeX More on the record page