Prompt Engineering Literacy Scale Dataset for Prompt Quality Assessment

A curated dataset for evaluating prompt engineering literacy. The dataset contains prompts from multiple domains annotated using the Prompt Engineering Literacy Scale (PELS), including six rubric scores, final weighted scores, and expert-written justifications. It is intended for research on prompt evaluation, LLM assessment, supervised fine-tuning, and Retrieval-Augmented Generation (RAG) systems.

About Dataset

The Prompt Engineering Literacy Scale (PELS) dataset is a human-annotated benchmark designed to evaluate the quality of prompts written for Large Language Models (LLMs). The dataset contains prompts collected across multiple application domains and evaluated using a structured six-dimensional rubric. Each sample includes: Prompt text Domain category Six rubric scores (C1–C6) Final weighted score Human-written justification The dataset supports research in: Prompt Engineering LLM Evaluation Prompt Quality Assessment Retrieval-Augmented Generation (RAG) AI Education Fine-tuning Language Models

Purpose of Dataset

The Purpose Of This Dataset Is To Support Research And Development In Prompt Engineering Evaluation By Providing A Human-annotated Benchmark For Assessing The Quality Of Prompts Written For Large Language Models (Llms). The Dataset Enables The Training, Fine-tuning, Benchmarking, And Evaluation Of Ai Systems That Automatically Score Prompts, Generate Feedback, And Improve Prompt Engineering Literacy. It Is Also Intended For Applications In Ai Education, Retrieval-augmented Generation (Rag), And Llm-based Assessment Systems.

Dataset Metadata

License

MIT

Geographical coverage

Sector

Education and Skill Development

Author

Rishi Phadale, Sneha Gadhari, Eshanika Amballa, Sheshank Singh, Trinetrini Quantum

Source Community

AIKosh Opensource Community

Uploaded by

Rishi Raj Phadale

Data Quality Score (Beta)

4.5

Dataset type

Structured

Frequency

Time Granularity

Year range

N.A.

Date & Time

07/07/26 11:09:50

Visibility

Open

Hosted / Redirected

Hosted

Data Type

Hybrid

Data Collection Method

The Dataset Was Created Through A Combination Of Manual Curation And Expert Annotation. Prompt Samples Were Collected From Publicly Available Prompt Repositories And Open-source Communities (Such As Opengpt, Hugging Face, And Similar Resources), Along With Original Human-written Prompts Created By The Project Team. The Collected Prompts Were Reviewed, Categorized Across Multiple Domains, And Evaluated Using The Prompt Engineering Literacy Scale (Pels) Rubric. Each Prompt Was Assigned Six Rubric Scores, A Final Weighted Score, And A Detailed Human-written Justification. The Curated Dataset Was Subsequently Used For Research On Prompt Evaluation, Retrieval-augmented Generation (Rag), And Llm Fine-tuning. The Dataset Was Created Through A Structured Pipeline Involving Prompt Collection From Public Repositories And Human Contributors, Manual Curation, Domain Classification, Expert Annotation Using The Six-dimensional Prompt Engineering Literacy Scale (Pels) Rubric, Quality Validation, And Consistency Checks. The Dataset Is Intended For Llm Evaluation, Prompt Assessment, Rag, And Ai Education.