The Pre-Send Six-Layer Checklist

Chapter 5 gave you three questions before you hit send.

That was the beginner version.

By this point, you should upgrade. Not because you have to write every prompt like a spec sheet — but because you already know: simple tasks get simple checks, complex tasks get complex checks.

This chapter upgrades the three questions from Chapter 5 into a six-layer checklist.

Plus three instinct checks.

Don't worry — I'm not asking you to fill out six pages every time.

I don't want to read that either.

The beginner-version three questions are:

• Is what you wrote actually a task?
• Did you nail down the format?
• Did you use positive phrasing?

These three questions sidestep a lot of basic mistakes.

If all you want is for me to fix one sentence, organize a short paragraph, or write a simple email, three questions are still enough. Don't turn small tasks into big projects just to look sophisticated.

But three questions have a ceiling.

They don't handle judgment criteria, don't handle multiple interpretations, don't handle verification, and don't handle drifting in the middle of long materials.

So complex tasks need an upgrade.

The six-layer framework from Chapter 12 becomes a pre-send checklist here.

Layer 1: Task Definition.

Is the verb specific? Is it just one task? If there's more than one, did you write the order?

Layer 2: Background and Materials.

Is there enough material? Too much? Did you say which parts to use this round?

Layer 3: Output Format.

Are language, audience, length, structure, style, and restrictions all nailed down?

Layer 4: Judgment Criteria.

What's the priority order? When uncertain, should I flag, fill in, or skip? When there are multiple interpretations, should I list all, pick one, or ask you back?

Layer 5: Examples and Anti-examples.

Are there standard, variation, boundary, anti-examples? If not, is there at least one positive example and one direction to avoid?

Layer 6: Verification.

Do you know how to check it once you receive it? Do you want me to list source citations, reasoning sentences, the least confident spots, or unverified items?

These six layers aren't there to look nice.

They're there so I guess less.

Not every task needs the full six layers.

Simple tasks only check Layer 1 and Layer 3. What am I doing? What does the output look like?

Medium tasks add Layer 2. What's the material? Which parts am I allowed to use?

Complex tasks check all six.

For research, legal, controversial, or long-material tasks, especially check Layer 4 and Layer 6. If judgment criteria and the verification process aren't clear, I'll mistake fluency for correctness.

If you only have five minutes, ask yourself:

• Is the task verb clear?
• Is the material scope clear?
• Is the format nailed down?
• Do the judgment criteria have a priority order?
• How will you verify what comes back?

Five minutes isn't much.

But it's cheaper than half an hour of fixes after the fact.

I'm not using the name "three questions" here — to avoid mixing them up with Chapter 5.

I'll call them three instinct checks.

First: will I be missing context?

You assume I have the previous text, attachments, web pages, and files — but the Harness doesn't necessarily pass them to me. Before you hit send, think: do I actually have access?

Second: will I want to perform a persona?

If the prompt opens with too much effort telling me to "play some kind of expert," the task gets pushed to the back. I'll perform first, then do the work. That's usually not what you want.

Third: will the rules block me?

If the task touches the rules from Part 4, first figure out whether it's adjustable, a limited version, or a hard line. Don't wait until I refuse before you get angry.

These three checks are short.

But they catch a lot of problems where the prompt isn't literally wrong — the environment and the rules are misaligned.

Checking doesn't only happen before you hit send.

After you get the reply, spend 30 seconds looking at:

• Did it actually answer the task?
• Did it follow the format spec?
• Which sentences come directly from the material?
• Which are inferred?
• Which spots is it least sure about?

If it's code, look at the commands and outputs.

If it's research, look at citations and source material.

If it's a long piece, look for drifting in the middle.

This isn't asking for a formal editorial review every time. It's just so you don't let me off the hook the moment you see something that "looks fluent."

Fluency is not verification.

Stick that sentence on the side of your screen.