275 observations of prompt decomposition using the Nyquist-Shannon spectral format for AI model training
The sinc-prompt dataset contains structured decompositions of natural language prompts into the 6-band sinc format, derived from the Nyquist-Shannon sampling theorem applied to specification space.
Each observation transforms a single raw prompt (1 sample) into a complete 6-band signal reconstruction, eliminating the aliasing (hallucination) that occurs when LLMs receive under-specified inputs.
The dataset spans 15 professional domains: software engineering, business strategy, legal, healthcare, finance, marketing, product management, data science, customer success, software architecture, DevOps/SRE, and human resources.
Research paper: DOI 10.5281/zenodo.19152668 | Source: github.com/mdalexandre/sinc-llm
Every sinc prompt follows this JSON schema:
{
"formula": "x(t) = \u03a3 x(nT) \u00b7 sinc((t - nT) / T)",
"T": "specification-axis",
"fragments": [
{"n": 0, "t": "PERSONA", "x": "Who should answer (role, expertise, experience level)"},
{"n": 1, "t": "CONTEXT", "x": "Situation, facts, background, environment"},
{"n": 2, "t": "DATA", "x": "Specific inputs, measurements, code, documents"},
{"n": 3, "t": "CONSTRAINTS", "x": "Rules, boundaries, success/fail criteria (42.7% of quality)"},
{"n": 4, "t": "FORMAT", "x": "Output structure, tables, lists, code blocks (26.3% of quality)"},
{"n": 5, "t": "TASK", "x": "The objective in 1-2 sentences"}
]
}| Band | Fragment | Weight | Description |
|---|---|---|---|
| n=3 | CONSTRAINTS | 42.7% | Rules, boundaries, criteria -- the most impactful band |
| n=4 | FORMAT | 26.3% | Output structure shapes response organization |
| n=0 | PERSONA | 7.0% | Role identity sets expertise framing |
| n=1 | CONTEXT | 6.3% | Situational facts ground the response |
| n=2 | DATA | 3.8% | Specific inputs reduce ambiguity |
| n=5 | TASK | 2.8% | Objective provides direction |
| # | Raw Prompt | Task Type | Domain | Bands |
|---|---|---|---|---|
| 1 | Review this Python function for bugs and performance issues | code_review | Software Engineering | 6/6 |
| 2 | My React app crashes when I click the submit button | bug_debugging | Software Engineering | 6/6 |
| 3 | Research the AI coding assistant market for me | market_research | Business Strategy | 6/6 |
| 4 | Review this SaaS contract for any issues | legal_contract_review | Legal | 6/6 |
| 5 | What could cause persistent fatigue and joint pain? | medical_diagnosis | Healthcare | 6/6 |
Showing 5 of 275 entries. View all 15 published examples with full sinc JSON.
One JSON object per line. Each object contains:
{
"raw_prompt": "string",
"sinc_json": { ... full 6-band sinc object ... },
"task_type": "string",
"domain": "string"
}15 examples published. Full 275-observation dataset available upon request.
If you use this dataset in research or training, please cite:
@misc{alexandre2026sinc,
author = {Alexandre, Mario},
title = {sinc-LLM: Nyquist-Shannon Prompt Decomposition for Large Language Models},
year = {2026},
publisher = {Zenodo},
doi = {10.5281/zenodo.19152668},
url = {https://doi.org/10.5281/zenodo.19152668},
note = {Dataset: 275 observations, 15 domains, 6-band spectral format}
}This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.
You are free to share and adapt the material for any purpose, including commercial use, provided you give appropriate credit.