Escape Conditions for Recursive Self-Training
Nine verification reports can represent nine independent observations or nine copies of one. This technical working paper measures how much fresh evidence a recursive learner receives, how much survives five training updates, and when trusting faulty source identifiers increases predictive error. Across 20 paired simulation trials, correct source identification retained a predictive advantage of 11.95 percentage points of baseline risk after five updates. Exact enumeration gives break-even case-corruption rates of 29.5% to 50.2% under two specified fault mechanisms. The eight-page paper includes three figures and reproducible Python experiments. I proudly acknowledge the role of human–AI collaboration in developing this work. Working with ChatGPT (OpenAI) has helped me expand my strategic thinking, examine competing explanations, and connect my proposals to relevant scientific and technical research. My experience, objectives, and judgment guide the work, and I retain responsibility for its claims and conclusions. I regard transparent, rigorous collaboration with AI as a constructive method for advancing ideas that may benefit humanity.
Authors
- Naveed Islam Butt
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-18
- DOI
- https://doi.org/10.5281/zenodo.24353373
- Primary Topic
- Adversarial Robustness in Machine Learning
- Type
- article
- Field-Weighted Citation Impact
- 0.00