The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
Delta Frame Validation Study (mirror)
Canonical record and citation: Zenodo, https://doi.org/10.5281/zenodo.22598205. This mirror exists for discoverability. If the two differ, Zenodo wins.
What it is
50 regulated-domain Python backend tasks (money, idempotency, date and time, authentication and access), each generated twice by five frontier models from five vendors: once from the bare prompt, once from the same prompt preceded by a 267-word specification frame. 500 outputs total, one attempt each, temperature 0 where the endpoint accepts it.
Included: the 100 prompts with SHA-256, the frame, nine AST-based checkers with validity fixtures, scoring and statistics code with seeds, Bandit results, the manual adjudication table, run and verifier logs, and the pre-registration with four dated addenda.
Headline result
Findings in nine named defect classes: 148 in the bare arm, 23 in the frame arm. Every model improved. None got worse. Bandit, run independently, moved from 53 medium-or-high issues to 11.
Files
One archive, byte-identical to the Zenodo deposit. Unzip and see README inside for the layout and the re-scoring instructions.
How to cite
Dhuri, S. (2026). Delta Frame Validation Study: complete results dataset (500 generations, 5 model families, pre-registered paired evaluation). Zenodo. https://doi.org/10.5281/zenodo.22598205
@misc{dhuri2026dataset,
author = {Dhuri, Sandeep},
title = {Delta Frame Validation Study: complete results dataset (500 generations, 5 model families, pre-registered paired evaluation)},
year = {2026},
publisher = {Zenodo},
doi = {10.5281/zenodo.22598205},
url = {https://doi.org/10.5281/zenodo.22598205}
}
Author: Sandeep Dhuri, ORCID 0009-0008-1829-2431. License: CC BY 4.0.
- Downloads last month
- 21