Lab 8 — Drafting and verifying technical prose with an AI assistant

Weeks 11–12 — a workflow walkthrough on a small Quarto YAML rendering issue

A hands-on walkthrough of the ask → record → verify → correct → disclose workflow you will use on the Week 11 AI debugging audit and the Week 12 drafting and critique work. The companion conceptual reading is A mental model for AI assistants and a verification habit. Lab 8 is shared across the AI module; you will return to it in Week 12.

You have written ten three-line AI Use Notes already. This lab is the first time the AI use itself is the substance of the walkthrough.

The lab walks a single small workflow on a single small illustrative sample — a Quarto YAML block with a deliberate typo that breaks numbering. The lab sample is illustrative only; the Week 11 work uses a different sample. What carries over is the workflow: ask the assistant, record what it says, verify the load-bearing claims, correct what needs correcting, disclose with a three-line AI Use Note.

This lab is tool-agnostic. ChatGPT, Claude, Copilot, Gemini, Cursor — any of these is fine. A free-tier account is sufficient. No paid subscription is required.

What you’ll need

VS Code with the Quarto extension installed and working (Lab 1).
A working Quarto + TinyTeX render chain (you have rendered PDFs in Weeks 1–10).
A browser tab open to an AI assistant of your choice.

You do not need any new R packages, any new LaTeX packages, or any new editor. The lab is short — a single sitting.

The illustrative sample

The sample for this lab is a small Quarto YAML block. Save the following as lab08.qmd in a lab08/ folder inside your math-software-portfolio/:

---
title: "A small report"
author: "Your Name"
format:
  pdf:
    toc: true
    number-section: true
geometry:
  - margin=1in
---

# Setup

This is a placeholder section. Lorem ipsum dolor sit amet.

# Results

This is another placeholder section.

# Discussion

This is a third placeholder section.

Render this document (Quarto: Preview, or quarto render lab08.qmd in a terminal). The render will complete without an error message. But the section numbers will not appear in the rendered PDF. Open the PDF and confirm: you see “Setup,” “Results,” “Discussion” as headings, but no “1”, “2”, “3” in front of them.

That is the bug for the lab. The render works, the document looks reasonable at a glance, but a specific YAML option is silently being ignored.

The lab is about using an AI assistant to figure out why and verifying the assistant’s claim against the artifact.

Step 1 — ask the assistant

Open your assistant of choice in a browser tab. Ask a scoped, specific question — not “fix this for me” or “rewrite this for me.” A scoped ask describes what you observed and asks for help diagnosing.

Two examples of the same ask, in different shapes. Use one of these or your own in the same shape — do not paste a prompt verbatim from a course page into the assistant; ask in your own words:

I have a Quarto document with number-section: true under format: pdf: in the YAML header. The document renders to PDF without errors, but the section numbers don’t appear. Can you help me figure out why?
In Quarto, what is the correct YAML key for turning on automatic numbering of sections in a PDF render? I have number-section: true but it doesn’t seem to do anything.

Paste the YAML block from above (or describe it briefly) so the assistant has the actual artifact to comment on.

Step 2 — record what the assistant says

Read the assistant’s response. Summarize it in your own words. Do not paste the full conversation into your lab document; this is a walkthrough, not a transcript.

A few common patterns you may see:

The assistant identifies the typo (number-section should be number-sections, plural) and explains that Quarto silently ignores unknown YAML keys.
The assistant gives a long, vague answer about Quarto numbering in general without spotting the typo.
The assistant suggests adding a different option (e.g., number-depth, section-numbering, or some invented key) that won’t actually fix the problem.
The assistant produces a confident-sounding explanation that includes one true and one fabricated detail.

You do not need to know in advance which pattern your assistant will produce. You audit whatever your assistant actually said. The verification chain in Step 3 is what supports the lab — not the AI’s response.

Step 3 — verify the load-bearing claims

This is the load-bearing step of the lab. Whatever the assistant said in Step 2, you verify the claims it made about the YAML key, the Quarto behavior, and the fix.

Two checks are sufficient for this sample; do all of them even if the assistant skipped one:

Check the Quarto YAML key documentation. Open the Quarto reference for PDF format options (quarto.org/docs/reference/formats/pdf.html or whatever the current canonical page is). Search the page for number-section (singular) and for number-sections (plural). One of them is the canonical option name; the other is not.
Apply the assistant’s suggested fix and re-render. Edit your lab08.qmd to apply whatever change the assistant proposed. Render. Open the PDF. Did the section numbers appear?
- If yes, the assistant’s fix worked. The verification chain entry is: “Applied the assistant’s fix (number-section: true → [whatever the assistant suggested]); re-rendered; section numbers now appear as 1 / 2 / 3 in the rendered PDF.”
- If no, the assistant’s suggested fix did not work. The verification chain entry names what you tried and what you saw.

Write each verification operation as a short entry. Each entry is verb + target + outcome — what you did, what you checked it against, and what you observed.

Step 4 — correct what needs correcting

If the assistant’s fix worked, you are done with the correction step. Show the corrected YAML block in your lab document, with a one-sentence note describing the change.

If the assistant’s fix did not work, identify the actual fix yourself (likely: change number-section to number-sections — plural — based on what you found in the Quarto reference). Apply the actual fix, re-render, and confirm that section numbers now appear. Show the corrected YAML block in your lab document.

Either way, the corrected version belongs in your lab document so a reader can see what changed.

Step 5 — disclose with an AI Use Note

The lab document closes with the standard three-line AI Use Note:

Tool: [assistant name and approximate version or date]

Purpose: [one sentence — what you asked the assistant to do]

Verification: [a short paragraph — name the verification operations you performed and what each one showed]

The Verification line for Week 11 is a short paragraph, not a single line. It summarizes the verification chain in the body. Three labeled lines, not four. “Verification” is still the third labeled line.

An example Verification line for this lab (yours will be shorter or longer depending on what your assistant said):

Verification: I checked the Quarto PDF reference page and found that the canonical key is number-sections (plural), not number-section. I changed the typo in my YAML, ran quarto render lab08.qmd, and opened the PDF. Section numbers now appear as 1, 2, 3 in front of Setup, Results, and Discussion.

Save the lab document. Render once. Open the PDF and read it as a stranger would.

What this lab prepares you for

This is the workflow you will apply to the Week 11 AI debugging audit. The Week 11 sample is different — it is not a Quarto YAML typo — but the workflow is the same:

ask the assistant a scoped, specific question;
record what the assistant says;
verify the load-bearing claims against stable external evidence;
correct what needs correcting;
disclose with a three-line AI Use Note where the Verification line is a short paragraph.

The exact Week 11 prompt and submission details live in the course LMS.

You will return to this lab in Week 12 for the drafting and critique work. The same workflow steps apply, with Week 12 emphasizing the draft + critique + revise half of the arc.

Common patterns you may encounter

The assistant identified the typo immediately. Lucky draw. Your verification chain is still the same — confirm the canonical YAML key in the Quarto reference, apply the fix, re-render, and observe the numbers appearing.
The assistant gave a confident wrong answer. Apply the assistant’s suggested change. Re-render. Observe that the numbers do not appear. That is your verification chain entry: I applied the change the assistant suggested; the document still rendered without section numbers. Then go find the correct option in the Quarto reference yourself.
The assistant mentioned a real Quarto option that doesn’t fix this particular bug (e.g., suggested number-depth: 3). Apply it; observe what happens; check the reference. The audit chain naturally records what was suggested, what was tried, and what happened.
The free-tier assistant ran out before the conversation finished. Note that in your lab document. Continue the verification yourself.
The assistant produced a long technical explanation about LaTeX \setcounter{secnumdepth}{...} machinery. This is possible. Quarto does not require manipulating the LaTeX preamble for section numbers; the canonical Quarto option is the right answer. The chain captures what you tried.

The point of the lab is that every load-bearing claim the assistant makes gets checked against an artifact you can see or a reference page you can read. The lab is not “did the AI get it right?” — it is “did I verify what the AI said?”

Looking ahead

The Week 11 work — your AI debugging audit — uses a different sample and applies the same workflow on a topic with externally checkable answers. The exact prompt and submission details live in the course LMS.

In Week 12 you will return to Lab 8 for the drafting-and- critique half of Module D. The same five-step workflow applies; Week 12 emphasizes drafting prose with an assistant and then critiquing what the assistant produced.

Then Week 13 brings the Portfolio/workflow conference, where your AI Use Notes — from Week 11, Week 12, and earlier weeks — are part of the portfolio review.