The Deterministic Horizon: Impossibility Results as Design Specifications for Trustworthy AI Systems

In three linesTheoretical paper proving architecture alone sets an accuracy ceiling (Deterministic Horizon) between 19–31 layers across 12 transformers. Beyond it, no training improves it. Converts 16 impossibility results (Turing, Arrow, No Free Lunch) into design rules for trustworthy AI systems, with computable bounds and quantified violation costs.

Summary generated by Claude — human-verified