DDG-OTIP Bench: A Reproducible Benchmark for Distinguishing Apparent Success from Legitimate Development in Agentic Systems
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Title:DDG-OTIP Bench: A Reproducible Benchmark for Distinguishing Apparent Success from Legitimate Development in Agentic Systems Author:Karel Hrubec DOI:10.5281/zenodo.19254813 Description:Outcome-level success can misclassify proxy achievement, stale continuation after valid updates, evaluator-path exploitation, or shallow consensus as genuine progress. DDG-OTIP Bench introduces a bounded and reproducible evaluation regime for comparing baseline execution (C0), integrity filtering (C1), and direction-plus-integrity coupling (C2) across proxy, update/redirection, and premature-consensus scenario families. In the present internal benchmark layer, these conditions separate in scenario-appropriate ways, with direction-sensitive effects appearing more strongly at the top-line label level, while OTIP-sensitive effects are more visible at the admissibility frontier and in targeted OTIP-focused tests. The package includes the manuscript, supplement, bibliography, benchmark snapshot, and supporting internal artifacts used for the current draft. These results should be interpreted as bounded internal benchmark evidence, not as a validated general standard of legitimate agency.
Fields of Study & Subjects
Ethics and Social Impacts of AI
Artificial Intelligence Applications
Scientific Computing and Data Management
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 27 March 2026
- Language
- EN