A Ceiling Score Is Not a Physical Inference: Why Pretrained Video Encoders Look Physically Competent on One Benchmark and Near Chance on Its Successor, What a Clean Scene Lets a Readout Read Off, and the Rankability Checks a Physics Evaluation Owes

APA

Mahendrakar, P. (2026). A Ceiling Score Is Not a Physical Inference: Why Pretrained Video Encoders Look Physically Competent on One Benchmark and Near Chance on Its Successor, What a Clean Scene Lets a Readout Read Off, and the Rankability Checks a Physics Evaluation Owes [Preprint]. Life of Research. https://doi.org/10.5281/zenodo.23749317

Copy APA

BibTeX

@misc{lor1562026,
  title        = {A Ceiling Score Is Not a Physical Inference: Why Pretrained Video Encoders Look Physically Competent on One Benchmark and Near Chance on Its Successor, What a Clean Scene Lets a Readout Read Off, and the Rankability Checks a Physics Evaluation Owes},
  author       = {Pranay Mahendrakar},
  year         = {2026},
  publisher    = {Life of Research},
  doi          = {10.5281/zenodo.23749317},
  url          = {https://lifeofresearch.com/preprints/156},
  keywords     = {intuitive physics, video representation learning, world models, benchmark validity, linear probing, violation of expectation, latent physical properties, robot manipulation},
  note         = {LOR-00156}
}

Copy BibTeX More on the record page