Lab 1 — LLM Context Mastery: Your Starting Profile

Points: 100 | Due: Tue May 19, 11:59 PM MT | Time: ~100 min hands-on | Submit: 3 Markdown files

Why context engineering matters (start here)

Analytics used to be "how well do you know Excel or Tableau." It is increasingly "how clearly can you describe what you want to an LLM, and what context can you give it." The tool fluency still matters. The leverage now sits in the layer above the tool — the role, the audience, the constraints, the source material — that you give the LLM before you ask it to produce anything. That layer is called context, and engineering it well is what this course teaches.

Concrete example. Two students paste the same prompt into the same LLM at the same moment:

"Summarize Alphabet's Q1 2026 10-Q in five bullets."

Student A — no context
Sends just the prompt. The LLM has no idea who is asking, why, or what filing they mean. It produces a generic five-bullet summary that may or may not match the actual Q1 2026 numbers — and may invent ones that look plausible.

Student B — full context
First uploads role_profile.md ("I am tracking toward in-house FP&A; I care about segment-level operating margin and capex commentary") and about_me.md ("I have written VLOOKUP and SUMIFS but not used Power Query"). Then attaches the actual 10-Q PDF and sends the same prompt. The LLM produces a summary anchored in the real filing, framed for someone in FP&A, with terms Student B can actually verify.

Same model. Same prompt. Different outcome — entirely because of context. Lab 1 builds the files Student B used. HW1 lets you experience the difference yourself in a controlled before/after. This is the load-bearing skill of the term — start here.

Want your LLM to read this whole lab? Paste this URL into your LLM session and it can fetch the full lab page:
https://seanmccaman.com/acctg5150/2026-summer/week-1/lab01_canvas_page.html
Most LLMs will follow links to read all three templates and the Example Managing-Up Document automatically. If yours does not, the direct markdown URLs are below.

Direct markdown URLs (paste these too if your LLM does not follow links)

https://raw.githubusercontent.com/sean-mccaman/acctg5150-090/main/week-01/Example_Managing_Up_Doc.md
https://raw.githubusercontent.com/sean-mccaman/acctg5150-090/main/week-01/lab01_spec_template.md
https://raw.githubusercontent.com/sean-mccaman/acctg5150-090/main/week-01/role_profile_template.md
https://raw.githubusercontent.com/sean-mccaman/acctg5150-090/main/week-01/about_me_template.md

What you'll do

Build three Markdown files. Each does a different job:

role_profile.md — describes the role construct you're tracking toward (communication norms, work standards, format identity, LLM directives). It's intentionally swappable; later weeks will provide alternate role_profile files so you can watch how the same prompt produces different LLM output.
about_me.md — describes you as a student: background, career picture, current skills, what you want to learn this term, the questions you're carrying, and a behavior table that shapes how the LLM tutors you. Uploaded to your LLM in HW1 and every later assignment.
lab01_spec.md — the spec-doc pattern you'll author for every lab. Defining what you want clearly is the skill that's taking over the accounting industry.

See it done first. Sean's filled-in pack: Lab 1 worked-example pack. Read for structure; write your own. Copying his content won't fit your role, your skills, or your goals — and graders will see it immediately.

The interaction protocol — open once, apply in Phases 3 and 4

Upload your source material to the LLM (Example Managing-Up Document, prior writing samples, this Canvas page).
Ask the LLM to interview you BEFORE drafting. Example: "Before you draft my role_profile.md, ask me five questions about how I want to communicate as a future auditor. Wait for my answers."
Answer in your own voice. One-word answers produce generic drafts.
Ask for a first draft in the template's structure.
Edit critically. Cut anything that sounds like a stranger wrote it.
Ask the LLM to review your edited version against the rubric and the Example Managing-Up Document Skill-Ready Checklist.
Iterate 2-3 rounds.
Run the "Validate it" forcing function at the end of the file (Section 6 of role_profile, Section 8 of about_me). The LLM asks you 5 questions. Iterate based on what surfaces.

Phase 1 — Environment Setup (~20 min)

Can't install VS Code? Read this first.

Three alternatives, in order of preference:

A. Notepad / TextEdit + browser viewer. Edit as plain text. Render at stackedit.io. On Mac, set TextEdit to Format → Make Plain Text first. Save with .md extension explicitly (in Notepad, change "Save as type" to "All Files" and type role_profile.md).

B. Google Docs with Markdown export. Draft in Docs, then File → Download → Markdown (.md). If your Workspace tenant disabled the export, install the Docs to Markdown add-on. Open the exported file in a viewer before submitting — Docs converts smart quotes and em-dashes in ways that confuse Markdown.

C. Pure LLM workspace. Draft inside ChatGPT/Claude, copy out into Notepad/TextEdit, save as .md. Most error-prone for beginners.

If using A, B, or C: skip Steps 1.1, 1.3, 1.4 below. Do Step 1.2, 1.5, and 1.6.

1.1 — Install VS Code

Download from code.visualstudio.com.
Run installer with default options.

1.2 — Create your folder

Make a folder named acctg5150_lab01_yourname on your Desktop or in Documents. (Replace yourname with your actual last name.)

1.3 — Open the folder in VS Code

File → Open Folder... (Ctrl+K Ctrl+O on Windows, Cmd+O on Mac). Select your folder. The Explorer panel on the left shows the folder name.

1.4 — Verify Markdown preview works

Create a file in Explorer named test.md.
Type # Hello on the first line. Save (Ctrl+S / Cmd+S).
Open side-by-side preview: Ctrl+K V (Windows) / Cmd+K V (Mac).
You should see "Hello" rendered as a heading on the right. Delete test.md when done.

If preview doesn't work: check the file ends in .md, or fall back to stackedit.io.

1.5 — Download the three templates

Click each button. Save into your folder. Then rename each file to drop the _template suffix.

lab01_spec_template.md role_profile_template.md about_me_template.md

You should end up with: lab01_spec.md, role_profile.md, about_me.md.

1.6 — Read the Example Managing-Up Document

Open Example Managing-Up Document and read end to end (~10 min). You'll apply its Skill-Ready Checklist to your own work in Phase 5.

Phase 1 done when: editor open, folder ready, preview works, three .md files in your folder, you've read the Example Managing-Up Document.

Phase 2 — Customize lab01_spec.md (sections 1-4) (~10 min)

Open lab01_spec.md. The template has the four canonical spec sections; most content is provided because Lab 1's deliverables (your own profile docs) make build-and-acceptance criteria a bit unusual. Your job: read end to end, then customize.

File manifest — confirm the three files listed match what you'll submit. Check the boxes when each file is done (in Phase 5).
Build Requirements + Review/Acceptance Criteria — read what's there. Add a line or two specific to your role if it sharpens the spec; otherwise leave as-is. Future labs require more authorship here.
How I would grade this submission — adjust criteria/weights to match what you think matters. Points must sum to exactly 100. Not 95, not 105.
Skip Section 5 for now — you'll fill in Self-Assessment in Phase 5.

Why spec-first. Defining what you want before producing it is the constraint-definition lesson — and the skill taking over the accounting industry as LLMs handle more execution. You'll author a spec doc for every lab. Note: your Phase 2 rubric is a hypothesis; you will revise it in Phase 5 once the work exists. That's the point.

Phase 2 done when: spec sections 1-4 customized, rubric sums to exactly 100, Section 5 still blank.

Phase 3 — Build role_profile.md (~25 min)

What this file is. role_profile.md describes a role construct, not you personally. You're building one for the route you're tracking toward — but the file itself is meant to be swappable. Later weeks will provide alternate role_profile files (auditor, tax specialist, etc.) so you can swap them in and watch how the same prompt produces different LLM output. Keep this file role-construct-only. Personal "why I picked it" content goes in about_me.md Section 2.

3.1 — Choose your accounting route

Pick one. This anchors the whole file.

Auditor — external audit, Big Four or regional firm
In-house accountant — corporate finance, controller, FP&A, internal audit
Tax specialist — public tax practice, corporate tax
Government accountant — federal, state, local, GAO, IRS
Other — name and describe what the role does and who it serves

You can change later. Pick one now. Note your career picture (the why) goes in about_me.md Section 2 — Section 1 of role_profile is just the role's job description.

3.2 — Open role_profile.md and apply the interaction protocol

Open the file. Apply the 8-step interaction protocol (top of page). Fill in all 5 content sections, then run Section 6 (the forcing function).

Structure when complete:

# Role Profile
## 1. Role overview            (route + 2-3 sentences on what this role does)
## 2. Communication norms      (audiences, document types, conventions)
## 3. What this role values    (work standards: citations, rigor, formatting)
## 4. Visual identity          (palette, fonts, deliverable conventions)
## 5. LLM directives           (how an LLM should respond serving this role)
## 6. Validate it              (forcing function — does this profile feel coherent?)

Section 5 (LLM directives) is the test. If your directives could apply to any role ("be helpful, be accurate"), the file isn't doing its job. They should sound role-specific — an auditor's directives shouldn't read interchangeably with a tax specialist's.

Section 6 — the forcing function. Once Sections 1-5 are drafted, paste the file back to your LLM with the prompt in Section 6. The LLM asks 5 questions about whether the profile is coherent and role-specific (vs generic). Iterate the file based on what surfaces.

Phase 3 done when: all 6 sections filled. The role description is specific enough that swapping it for another role would noticeably change LLM output. You've answered the LLM's questions and made at least one edit. Save the file.

Phase 4 — Build about_me.md (~35 min)

Open about_me.md. Apply the interaction protocol. This is the bigger of the two profile files — it covers your current identity (Sections 1, 3, 4), your future-facing picture (Sections 2, 5, 6), the tutoring contract you give your LLM (Section 7), and a forcing function that tests the whole thing (Section 8).

Structure:

# About Me
## 1. Background and experiences        (3-6 sentences)
## 2. Career picture (3-5 year horizon) (target role, employer, credentials, uncertainty)
## 3. Current technical skills          (per tool: 1-3 specific things, OR "never used")
## 4. How I learn and what makes concepts stick  (3-5 sentences)
## 5. What I want to learn this term    (1-3 capabilities, anchored to Section 3)
## 6. Questions I would like answered   (2-4 questions)
## 7. What I want this file to help with (behavior table — your tutoring contract)
## 8. Validate it                       (forcing function)

Section 2 — Career picture. Sketch where you see yourself 3-5 years out: target role, target employer, credentials, what you're still figuring out. One sentence linking back to the role you picked in role_profile.md Section 1. Honest uncertainty is fine; vagueness ("I want to be successful") is not. The instructor uses this section later to check whether your work and learning over the term align with the picture you declared.

Section 3 must be concrete. "I'm comfortable with Excel" tells a future LLM nothing. "I can build a pivot table without looking up the steps; I've written VLOOKUP and SUMIFS; I have not used Power Query" is something the LLM can act on. For tools you've never touched: write "never used."

Section 5 must reference Section 3. Your end-of-term goals only make sense relative to where you start. Bad: "Get better at Python." Good: "Build a script that loads a CSV, cleans missing values, and produces a chart — without searching for syntax. (Today: I have not used Python.)"

Section 7 — Behavior table. This is the row-by-row contract you give your LLM for how to help you. Use the template's starter rows AND add 2-3 of your own. The LLM follows these instructions every time you upload this file. The verification row ("Get a number or fact from the LLM → Show its work and tell me how to verify it against the source") is the load-bearing one for HW1 and beyond — keep it concrete.

Section 8 — the forcing function. Once Sections 1-7 are drafted, paste the file back to your LLM with the prompt in Section 8. The LLM asks 5 questions about your skill claims, career picture, learning goals, the questions you're carrying, and your behavior table entries. Iterate the file based on what surfaces.

Phase 4 done when: all 8 sections filled. Skills are concrete; Section 5 goals reference Section 3; Section 7 has at least 5 rows including a verification row; Section 2 career picture is concrete enough to grade alignment against later. You've answered the LLM's questions and made at least one edit. Save the file.

Phase 5 — Finalize spec doc + submit (~15 min)

5.1 — Reopen lab01_spec.md and complete Section 5 (Self-Assessment)

Fill in the Submission Gate Checklist (SG-01 through SG-05). For each item, mark [OK], [WARN], or [FIX] with one sentence of evidence. Fix any [FIX] items before submitting.

5.2 — Reconcile your rubric

The work shifted while you drafted (it always does). Confirm your Section 4 rubric still describes what you actually produced. Adjust weights if needed — must still sum to exactly 100.

5.3 — Run the Skill-Ready Checklist

Run your work through the Skill-Ready Checklist in the Example Managing-Up Document — sections "Applies to Every Deliverable" and "Written Analysis and Reports." Note any [WARN] items in the spec doc Self-Assessment with one sentence of evidence.

5.4 — Submit

Click Submit Assignment at the top of this Canvas page.
Use Add Another File to upload all three files.
Confirm filenames exactly match: lab01_spec.md, role_profile.md, about_me.md.
Click Submit.

Format strictness. Markdown only (.md). PDFs, Word docs, or Google Doc links count as a missed format requirement.

Next: HW1. HW1 (Alphabet Q1 2026 10-Q Investigation) uses role_profile.md and about_me.md as Pass 2 context. Do Lab 1 first.

Grading rubric

Total: 100 points (lab work 85 + spec doc 15).

Lab work (85 pts)

Category	Pts	Criteria
Completeness	25	All three files present, named correctly, every required section filled (including the Section 6 / Section 8 forcing-function follow-up).
Specificity	25	Concrete, student-specific details rather than generic statements. Behavior table in about_me Section 7 is usable, not aspirational.
Reusability	22	Concise, organized, useful as future LLM context.
Professional polish	13	Clear, proofread, well formatted, free of clutter.

Spec doc (15 pts)

Category	Pts	Criteria
File manifest accuracy	5	Manifest lists every required file; every file is Markdown; every file submitted.
Self-assessment quality	10	Usable rubric in your own words (sums to exactly 100); self-rating with one sentence of evidence per criterion.

Format requirements

Markdown only (.md). PDFs, Word docs, Google Doc links count as missed format.
Use # and ## headings as the templates specify.
Concise. If a section runs longer than half a page, ask whether a reader would actually use it.
No private credentials, tokens, student records, grades, or sensitive information.
Proofread.

Need help?

Preview broken? Use stackedit.io as backup.
Stuck on the role? Pick the one closest to your current internship, prior coursework, or the one you're most curious about. You can revise later.
AI policy: see syllabus. For Lab 1, use any LLM available to you.
Questions: Canvas Inbox message preferred over email.

Last updated: 2026-05-12.