The 6 Frequency Bands

Every prompt is a signal with 6 frequency components. Missing any band forces the model to hallucinate the gap.

Persona

7.0%

Who should answer. Role identity and domain expertise.

Context

6.3%

Situation, facts, dates, and background information.

Data

3.8%

Specific inputs, metrics, numbers, and measurements.

Constraints

42.7%

Rules, limits, and requirements. The dominant frequency band.

Format

26.3%

Output structure: tables, lists, headers, code blocks.

Task

2.8%

The objective. Only 2.8% of quality despite being everyone's focus.

Aliasing vs Faithful Reconstruction

A raw prompt is an undersampled signal. The model hallucinates the missing bands, producing aliased output that looks plausible but contains fabricated information.

Before: 1-band (aliased)

Raw prompt. Model invents missing bands. Inconsistent, hedging, hallucinated output.

Clean signal from fully-sampled sinc prompt

After: 6-band (reconstructed)

sinc-prompt. All bands specified. Precise, direct, faithful output.

Custom Prompt Engineering

You are not paying for a prompt. You are paying for the output quality that changes your bottom line.

295x

Average ROI. A $197 prompt saves $58,200/year in API costs for a team spending $5K/month.

45 min

Daily time saved. No more re-prompting because the output was vague, hedged, or wrong format.

42.7%

Of output quality comes from CONSTRAINTS. Most prompts have zero. I write the ones your domain needs.

Proof

Every delivery includes a before/after comparison. You see the difference in 5 seconds. Not a promise. Evidence.

Who is this for?

Developers spending $1K+/month on APIs

Your prompts are structurally impoverished. Adding 39 tokens of constraints cuts costs by 97% because the model stops generating filler.

Teams getting inconsistent AI output

The model is not unreliable. Your prompt is underspecified. A sinc prompt locks the behavioral boundaries so every run produces the same structure.

Businesses where AI output affects revenue

Legal review, medical analysis, financial reporting, customer support. When the AI hedges, a human has to redo the work. The sinc prompt eliminates that rework.