Published August 13, 2026 | Version v1

Replication Package for "Prompt Complexity and Diminishing Returns in Generative AI Programming Assistance"

Description

This record contains the experimental dataset and replication materials supporting the study “Prompt Complexity and Diminishing Returns in Generative AI Programming Assistance.” The experiment compared seven prompting strategies—zero-shot, role-based, structured, few-shot, hint-based, Socratic, and iterative refinement—across six introductory Python tasks. The design comprised 42 task–strategy cells with five stochastic replications per cell, yielding 210 experimental episodes and 330 recorded AI responses. The deposit includes the scored dataset, frozen hidden tests and rubrics, raw prompt/response transcripts, offline token-count data, and scripts used to reproduce the blocked statistical analyses. Token counts are offline counts of recorded user-input and AI-output text and are not platform-reported API usage.

Files

jamovi_token_efficiency_210.csv

Files (680.8 kB)

Name Size
md5:612b066879ac3487b823b7d94dcf5a06
16.1 kB Preview Download
md5:f5d2647f90a6451cea6bf39a5b702686
358.9 kB Preview Download
md5:b66d71dd70d40a658e194aa8cd7b1db6
298.6 kB Download
md5:43aad833201af2d893d3a4e7663a6e1e
7.1 kB Preview Download