What a Clean Signal Actually Looks Like

What a Dirty Signal Looks Like

Here is a dirty signal: "Build me a pricing calculator." That is seven words. That is zero specification. The model then has to guess the programming language, the framework, the pricing model, the input format, the output format, the edge cases, the error handling, the styling, and the architecture. Each one of those guesses is a gap. Each gap is a chance for the model to get it wrong.

The model will make a pricing calculator. It will probably work, at a basic level. But it will show the model's guesses, not my real needs. My needs and the model's guesses almost never match.

Cleaning the Signal

A clean signal starts with the role. Who is doing this work. Not "you are a helpful assistant." That says nothing useful. "You are a senior Python engineer specializing in financial calculations with experience in real estate pricing models." That is a real role. It tells the model what domain to work in and what level of skill to use.

Then the context. What already exists. What system this connects to. What has been decided. What has not been decided yet. The model cannot see my project. It cannot read my code. It cannot see my past conversations unless I paste them in. Everything the model needs to know must go into the prompt.

Then the data. Specific inputs. Real examples. Reference material. The actual numbers, formats, and structures the output must use. Not descriptions of data. The data itself.

Then the constraints. What the output must do. What it must not do. How fast it must run. How it must look. What it must work with. Constraints are not limits that hold the model back. They are fences that keep it focused on the right answer.

Then the format. How the output should look. File format. Variable names. Comment style. Documentation style. If I do not say how to format it, the model picks for itself. It never picks what I want.

Then the task. The exact action. Not "build a calculator." More like: "implement a function that accepts a float, computes the square root, multiplies by the zone coefficient from the config table, and returns the base price rounded to two decimal places."

The Difference Is Everything

A dirty signal gets a response. A clean signal gets the right response. The gap between those two is huge. Think of asking a contractor "build me something" versus handing them blueprints with exact dimensions, materials, tolerances, and a timeline. The contractor can start either way. But only one of those leads to what you actually want.

I write clean signals now because dirty ones waste my time. Every minute I spend making a signal clear saves me ten minutes of fixing, rewriting, and trying again. The math is not even close.