First experiment

For people who have never published anything technical. Two sessions of up to two hours each, with no cloud account.

Session 1: from question to evidence

1. Create the experiment folder.

   npx -y moretes-builder-workshops experiment my-experiment --lang en

   It comes with PROPOSAL.md, EVIDENCE.md, ARTICLE.md and CHECKLIST.md.

2. Write the real question in PROPOSAL.md. Something that genuinely confused you. “I want to talk about SQS” is not a question yet. “Why didn’t more consumers shrink my backlog?” is.

3. Write the hypothesis before testing. It’s what turns a tutorial into learning: afterwards you’ll be able to say what changed in your head.

4. Define the smallest possible test. If it doesn’t fit in two hours, cut it down. Prefer running locally, with fictional data.

5. Run it and note it down right away in EVIDENCE.md. The command, what you expected, what happened. An error is a result too.

No topic idea?

Open the terminal and make a prediction: queue --arrivals 100 --service-ms 200 --consumers 20 --backlog 6000 --lang en. What do you think will come out? The queue capacity page takes it further.

Session 2: from evidence to article

1. Write the first draft yourself, from the evidence log. It can be ugly. Tell what you did, the way you would tell a colleague.

2. Ask for a review, not for writing. With the skill installed, ask something like this:

   Review my article with the builder-workshops skill. Don’t rewrite it. Point out what another person couldn’t repeat and what I claim without evidence.

3. Go through CHECKLIST.md. Open every link you cite, state the cost and say how to clean up what was created.

4. Publish and share. On your blog, on DEV.to, on AWS Builder Center or wherever you prefer.

Why AI only reviews

Three practical reasons:

• Your experience is the article’s value. A model didn’t test what you tested.
• You answer for the text. You need to be able to explain every claim when someone asks.
• It’s how you learn. Writing forces you to notice what you still don’t understand.

If AI helped, say with what: “clarity review” is enough.

Before publishing, check

• Another person can repeat it using only what’s written.
• Every number came from your test or has a source.
• The article says what the test does not prove.
• No key, account or customer data appears in text or images.
• Nothing in it is forbidden by your job or by a nondisclosure agreement.

See a finished example

The complete example shows the four stages with real outputs, including the two hypotheses that fell.