FPL Token Study: Raw API Outputs

FPL Token Study Beyond Syntax: Token-Level Evidence for a New Generation of Human–LLM Programming Languages Author: Tanjim Khan Nokib Institution: Karl-Franzens-Universität Graz, Computational Social Systems, Austria We conducted a controlled quantitative experiment varying five prompt design variables across 500 HumanEval prompts submitted to LLaMA 3.3 70B via the Groq API (temperature = 0.0, 2,500 total API calls), measuring five token-level output metrics. Key finding: Few-Shot prompts with programming-heavy vocabulary produced the lowest output entropy (M = 4.326 bits), translating into five design axioms for a Future Programmer Language (FPL).