Rho-Perfect: Correlation Ceiling For Subjective Evaluation Datasets
Fredrik Cumlin · 2026
Read Rho-Perfect: Correlation Ceiling For Subjective Evaluation Datasets by Fredrik Cumlin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-08552.txt.
Description
Subjective ratings are noisy, so even a perfect model cannot correlate perfectly with human judges, yet that ceiling is rarely measured. Fredrik Cumlin proposes rho-Perfect, an estimate of the highest correlation a model can achieve on subjectively rated data. His arXiv paper derives the estimate under heteroscedastic noise, the common case where noise varies across items.
Squaring rho-Perfect estimates test-retest correlation, which validates the approach. A speech-quality demonstration shows how the measure separates model shortcomings from data-quality problems. Readers get a practical yardstick: before blaming a model, check whether the ratings themselves cap its score.
Catalog record
| CREATOR | Fredrik Cumlin |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 21 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-08552