The Verification Habit

Why this matters

Verification is not optional cleanup. It is part of the work. For any output you will actually use: check claims that matter against sources or your own knowledge, test edge cases or numbers that would be expensive if wrong, read the output for confident fabrication, and decide deliberately what you are willing to own.

What you will be able to decide after this

• Whether an output is safe to use — or which parts are not.
• When the verification cost makes the task net negative.
• What you are willing to own, and what you are not.

Core lesson

The habit is four checks, applied before anything is used:

• Check claims that matter against sources or your own knowledge.
• Test edge cases or numbers that would be expensive if wrong.
• Read for confident fabrication — the fluent sentence that invented a fact.
• Decide deliberately what you are willing to own.

If you cannot verify it in reasonable time, the net value is probably negative. Stop or change the task.

Example verification prompt (after you have draft output)

Here is the draft output I received:

[paste output]

My original success criteria were:
[paste criteria]

Do the following:
1. List every factual claim or number in the draft.
2. Flag any claim that is not directly supported by the context I gave you.
3. Flag any place the draft softens or omits a risk I mentioned.
4. Tell me exactly what I still need to check myself before I can own this.

The model is not the verifier. This prompt structures the work; the checking is still yours. The last item — "what I still need to check myself" — is the point.

Worked example

A draft project update says the payment integration is "on track for a mid-quarter release." The context provided says integration is the main risk and has no confirmed date. The verification prompt flags the claim as unsupported, and the omission of the risk. The correct edit is not to soften the sentence — it is to state that the integration has no confirmed date and is the main risk. The verification caught the gap; the human made the fix.

Practice

Run the verification prompt on one draft output you already have. Keep the model's flags. Then independently check the top three flags yourself. Note any the model missed.

Apply — produce the artifact

Using the output from Module 2, produce a verified version. Document exactly what you checked, what you changed, and what you still cannot fully verify.

Verify

Have someone else look at your verified version and your check list. If they find an important claim you did not examine, fix the gap.

Sources

This module is original practice guidance based on the authoring standard and does not depend on a specific external factual claim. Editorial review is still required.