A Lift Is Not a Selection Gradient: What the Amplification-Lift Gap in Thinking Models Measures, Why an Answer-Aware Label and a Pooled Contrast Account for Much of It, and the Within-Prompt Test That Would Show Training Is Misdirected

APA

Mahendrakar, P. (2026). A Lift Is Not a Selection Gradient: What the Amplification-Lift Gap in Thinking Models Measures, Why an Answer-Aware Label and a Pooled Contrast Account for Much of It, and the Within-Prompt Test That Would Show Training Is Misdirected [Preprint]. Life of Research. https://doi.org/10.5281/zenodo.23951230

Copy APA

BibTeX

@misc{lor1872026,
  title        = {A Lift Is Not a Selection Gradient: What the Amplification-Lift Gap in Thinking Models Measures, Why an Answer-Aware Label and a Pooled Contrast Account for Much of It, and the Within-Prompt Test That Would Show Training Is Misdirected},
  author       = {Pranay Mahendrakar},
  year         = {2026},
  publisher    = {Life of Research},
  doi          = {10.5281/zenodo.23951230},
  url          = {https://lifeofresearch.com/preprints/187},
  keywords     = {reasoning models, chain of thought, reinforcement learning from verifiable rewards, self-correction, confidence calibration, process supervision, selection differential, evaluation methodology},
  note         = {LOR-00187}
}

Copy BibTeX More on the record page