Week 16 · Study guide

Final Exam Study Guide · Weeks 1–15 (Objectives 1–8)

Using Artificial Intelligence · AI 101 Fall 2026 · Prof. Quinn Fictional sample

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
This is a student-facing review guide. Read it, work the fresh practice, and follow the dated plan. Then run the paired Exam-Prep Tutorial (N) and take the Practice Final (O) for active recall. (This guide points to both — it does not repeat them.)
Also use this to plan your Capstone. Each objective section notes how the capstone integrates those concepts.

Integrity note for students. Every practice item in this guide is a fresh variant — new scenarios, new wording — with a vetted answer. None of these are the live final questions. Working them builds the skill the Final tests, honestly.

What the Final Covers (Read This First)


Exam	Final — cumulative, Weeks 1–15, all 8 Objectives
Format	25 items, 100 points (4 points each). A mix of concept, scenario, and matching items. Expect multiple-choice, four matching items (Cowork terms, structured-prompt components, modality-to-tool, and data-type-to-handling-rule), and a few true/false. Includes a "prompting fix" scenario item — read a weak prompt and choose the best revision.
Coverage (where the points are)	Obj 1 = 3 items (AI concepts & limits) · Obj 2 = 5 (the prompting arc) · Obj 3 = 2 (modalities & tools) · Obj 4 = 2 (verification) · Obj 5 = 4 (Cowork: agents, projects, skills, connectors, artifacts) · Obj 6 = 4 (automation: scheduled tasks, dispatch, computer use, Chrome, safety) · Obj 7 = 2 (responsible AI) · Obj 8 = 3 (capstone integration). The back half — Objectives 5–8 — carries 13 of 25 items, so budget the most time there (the midterm already covered Obj 1–4).
Weight	The Final is 25% of your course grade — the single largest assessment.
When it opens / where	Opens in the Week 16 module (Mon Dec 14); the window is open for six days. No quiz, discussion, assignment, or Studio in Week 16. AI is not permitted on the Final.
What to bring	Yourself, rested, and your understanding. The exam tests recall and application — no notes are needed in advance if you have worked this guide actively.

How to use this guide. Each objective section has four parts: (A) key ideas in plain language, (B) definitions and vocabulary to lock, (C) predictable mistakes and their cures, (D) capstone tie-in. After all eight objectives come fresh practice items and a dated study plan.

Objective 1 — What Generative AI Is & Its Limits (Weeks 1–2) · 3 items

(A) Key ideas, plain language

AI assistants generate text by predicting what token (word piece) is most plausible next, based on statistical patterns in training data. They do not understand, think, search the web (typically), or retrieve verified facts. The context window is finite — a very long conversation can push out earlier instructions. AI is confidently wrong — it sounds fluent whether it is right or completely fabricating.

(B) Vocabulary to lock

LLM = large language model; generates text by next-token prediction.
Token = the unit the model processes (roughly a word or word-fragment).
Training = the process of learning patterns from a large text dataset; has a cutoff date (no knowledge of events after the cutoff).
Context window = the total amount of text the model can hold at once; when exceeded, earlier content drops out.
Hallucination = generating confident, plausible-sounding text that is factually wrong or entirely fabricated.
Sycophancy = agreeing with or praising the user even when the user is wrong.
Turing test (Alan Turing, 1950, "Computing Machinery and Intelligence") = a test of whether a machine's text is indistinguishable from a human's in a conversation — it does NOT prove understanding or consciousness.
AI vs. search engine = a search engine indexes and retrieves; an LLM generates.

(C) Predictable mistakes → cures

❌ "The AI understands what I mean." → ✅ It predicts plausible next tokens — no understanding, no meaning.
❌ "Confident tone = accurate content." → ✅ The model generates fluent text even while fabricating — tone is irrelevant to accuracy.
❌ "A bigger context window makes the AI smarter or more truthful." → ✅ A larger window holds more, but does not reduce hallucination.
❌ "The Turing test proves a machine is conscious." → ✅ It tests conversational indistinguishability only.

(D) Capstone tie-in

Your capstone workflow relies on an AI that generates plausible-sounding content. Understanding next-token prediction and hallucination is why the verification step is required: the AI is not looking up your calendar events from a database — it is generating what sounds right, which can produce invented meetings, wrong times, or made-up names.

Objective 2 — Effective Prompting (Weeks 3–6) · 5 items

(A) Key ideas, plain language

The prompting arc runs from the basics of conversation to structured, reusable templates. The goal is to give the AI exactly enough information — no more, no less — to produce what you need without guessing, drifting, or praising you when you want criticism.

(B) Vocabulary to lock

Skill 1 — conversation: ask for guidance, iterate, counter sycophancy (explicitly ask for critique).
Skill 2/3 — provide content + emphasis: paste your notes or document; use Markdown (bold, headers), XML-style tags (<focus>...</focus>), and CAPS for must-dos to signal what matters.
Skill 4 — meta-prompting: "Ask me clarifying questions one at a time, then return a structured Markdown prompt." The AI helps write the prompt.
Skill 5 — structured-prompt components (nine): Context · Role · Goal · Audience · Constraints · Voice/Format · Data/Logic · Examples · Evaluation. Memorize: CRG-A-CVEIE or just the names in order.
Skill 6 — zero/one/few-shot: zero = no examples; one = exactly one; few = several examples (not "exactly one" — that is a classic misconception).
Skill 7 — simulation: difficult-conversation practice, pre-mortem, decision role-play, adaptive tutor. Critical rule: AI-generated quotes from simulated historical figures are generated text — never cite them as real.

(C) Predictable mistakes → cures

❌ "More words = a better prompt." → ✅ Clarity and the right components beat length.
❌ "Assigning a 'role' makes the AI actually expert/accurate." → ✅ A role shapes tone and framing — it does not grant real expertise or reduce hallucination.
❌ "Few-shot means exactly one example." → ✅ One example = one-shot; few-shot = several.
❌ "A simulated historical figure's 'quotes' are real." → ✅ Generated text — never cite as real, never present as the person's actual words.
❌ "Politeness improves output." → ✅ Structural components (Role, Constraints, Examples) improve output; politeness does not.

(D) Capstone tie-in

Your capstone's project instructions are a structured prompt. If the output is hallucinating or drifting, the fix is almost always in the prompt — add a constraint, clarify the goal, or give an example of the correct output format. Use Skill 5 to diagnose what is missing; use Skill 6 to show the AI the right output shape.

Objective 3 — Modalities & Choosing the Right Tool (Weeks 7, 9) · 2 items

(A) Key ideas, plain language

AI works across multiple input/output types — text, voice, audio, images, documents, video — and different tools are designed for different modalities. No single chatbot does everything equally well.

(B) Vocabulary to lock

Voice prompting = speak instead of type; the chatbot transcribes and responds.
Record → transcribe → analyze = the workflow for audio meetings: (1) record with any recorder; (2) transcribe with a transcription tool; (3) send transcript to a chatbot for summary and action items. Catch transcription errors and summary fabrications.
Image-to-text / handwriting recognition = multimodal AI analyzes a photo of handwritten notes.
Image generation = DALL·E (OpenAI), Midjourney, Adobe Firefly — separate category from chatbots.
Tool landscape (from W9):
Chatbots: ChatGPT, Claude, Gemini, Copilot, Grok, Meta AI
Image: DALL·E, Midjourney, Adobe Firefly, Google Imagen, Stable Diffusion
Audio/music: Suno, Udio (music); ElevenLabs (voice)
Video: Sora (OpenAI), Google Veo, Runway, Pika
Research: NotebookLM (Google), Perplexity
Coding: Claude Code, GitHub Copilot, Cursor

(C) Predictable mistakes → cures

❌ "Chatbots are text-only." → ✅ Modern chatbots handle voice, images, documents, and PDFs.
❌ "One chatbot does everything equally well." → ✅ Match the tool to the modality and task.
❌ "Transcription is always accurate." → ✅ Transcription tools make errors; always review before trusting the transcript.

(D) Capstone tie-in

If your capstone involves processing audio (meeting notes, interviews), use the record→transcribe→analyze workflow and catch transcription errors before they propagate into the summary.

Objective 4 — Verification & Critical Thinking (Week 10) · 2 items

(A) Key ideas, plain language

AI generates plausible text, not verified truth. Verification is the habit of checking AI output against external authoritative sources before trusting or acting on it.

(B) Vocabulary to lock

Hallucination shapes: invented citations (most common on the exam), fabricated statistics, fake case law, wrong arithmetic, fabricated quotes.
Sycophancy: the AI agrees with the user even when wrong; counter by explicitly asking for critique.
Verification workflow (four steps):
1. Ask for sources and check them in a library database or authoritative site.
2. Cross-check the claim in a second model or independent external source.
3. Ask the AI to critique itself: "Where might this be wrong?"
4. Request hedging: "If you're not certain, say so."
Why asking the same AI to check itself doesn't work: the model has the same biases that produced the error.
Why two AIs agreeing isn't proof: both may have learned the same wrong pattern.

(C) Predictable mistakes → cures

❌ "If the AI provides a citation, the source is real." → ✅ Invented citations are the most common and dangerous hallucination shape — always check.
❌ "Asking the AI to check itself is sufficient." → ✅ The model cannot reliably verify its own errors; check externally.
❌ "Confident tone = reliable." → ✅ AI sounds fluent whether fabricating or not.

(D) Capstone tie-in

The verification step in your capstone is the direct application of Obj 4. Document what error you found, what hallucination shape it was, and how you fixed or constrained it.

Objective 5 — Claude Cowork: Agents, Projects, Files, Skills, Connectors, Artifacts (Weeks 11–12) · 4 items

(A) Key ideas, plain language

Claude Cowork is the desktop automation platform where Claude acts as an agent — taking multi-step actions on your behalf — rather than just answering questions. The vocabulary in this objective is the source of the most Final items and the most Final misconceptions.

(B) Vocabulary to lock (the four-way distinction is the top exam target)

Term	Plain-language definition
Agent	Software that takes multi-step actions on your behalf (reads files, writes outputs, executes tasks) — vs. a chatbot, which replies in one turn
Project	A persistent, self-contained workspace with its own files, instructions, and memory that carries context across multiple tasks
Connected folder	A local folder on your computer that Claude can read from and write to
Skill	A reusable instruction set stored as a SKILL.md file that teaches Claude how to perform a specific type of task; built-in skills include docx, pptx, xlsx, pdf
Connector	A link to an external app using MCP (Model Context Protocol) — an open standard created by Anthropic — running only with the permissions you grant
Live artifact	A persistent, interactive view that refreshes with current data from your connected apps each time you open it (NOT a static file)
Plugin	A bundle of skills + connectors + sub-agents packaged as one installable unit

(C) Predictable mistakes → cures

❌ "An agent is just a chatbot." → ✅ An agent takes multi-step actions; a chatbot replies once. Fundamental difference.
❌ "A skill and a connector are the same thing." → ✅ A skill (SKILL.md) is a local instruction set; a connector is an MCP link to an external app. To pull Gmail data, you need the Gmail connector — a SKILL.md file cannot connect to an external app.
❌ "A live artifact is a static file." → ✅ It refreshes from your connected data every time you open it.
❌ "MCP is proprietary to Anthropic." → ✅ MCP is an open standard — other AI systems can implement it.
❌ "A plugin is the same as a single skill." → ✅ A plugin is a bundle of skills + connectors + sub-agents.

(D) Capstone tie-in

Your capstone demonstrates Obj 5 directly: which type of Cowork feature did you use? A project with a connected folder + a task is the minimum. Adding a connector (to pull real data), a skill (to format the output), or a live artifact (to display it) is where the Obj 5 depth shows.

Objective 6 — Automation: Scheduled Tasks, Dispatch, Computer Use, Chrome, Excel, Safety (Weeks 13–14) · 4 items

(A) Key ideas, plain language

Cowork can automate real tasks — schedule them, run them in the background, and even control your computer. Every one of these capabilities comes with a clear safety envelope: you cannot automate money movement, you must approve before irreversible actions, and scheduled tasks only run when your machine is awake.

(B) Vocabulary to lock

Feature	Key fact(s)
Scheduled task	Set a task to run automatically (recurring or one-time); use /schedule command or the Scheduled sidebar. CRITICAL: runs ONLY while computer is awake AND Claude desktop app is open. Skips if machine is asleep; re-runs on wake.
Dispatch	Start work to run autonomously in the background; you receive a notification with the result. Asynchronous — you are NOT watching each step. NOT a regular interactive chat.
Computer use	Claude controls native desktop applications via screenshots, clicks, and keyboard input. Requires permission; broad scope.
Claude in Chrome	Claude navigates and interacts with Chrome browser tabs. Available in beta on paid plans. Primary risk: prompt injection (malicious instructions hidden in web page content redirect Claude's actions). Defend with approval checkpoints, start on trusted sites, stop if Claude does something you didn't request.
Claude in Excel	Claude in a sidebar inside Microsoft Excel; reads, analyzes, modifies, and creates workbooks. Available in beta on paid plans.
The money rule (absolute)	You — not any AI agent — execute financial transactions, trades, and purchases. No exceptions. Financial sites are blocked by default in Claude in Chrome.
Least privilege	Connect only the folder, connector, or permission the task actually needs — no more.
Approval checkpoints	Add a human review step before any irreversible action (submitting a form, sending an email, deleting a file).

(C) Predictable mistakes → cures

❌ "Scheduled tasks run even if the computer is off." → ✅ Tasks require computer AWAKE + Claude app OPEN. The most heavily tested Obj 6 fact.
❌ "Computer use and a connector are the same." → ✅ Computer use = visual control of any desktop app; connector = API link to a specific service.
❌ "It's fine to let an agent move money if I give permission." → ✅ The money rule is absolute — no agent executes financial transactions on your behalf.
❌ "The AI filters all prompt-injection attempts automatically." → ✅ AI filters reduce risk but are "not a security boundary" (per official docs) — always use approval checkpoints.
❌ "Dispatch runs on Anthropic's cloud so the computer can be off." → ✅ Dispatch still runs on your desktop; the computer must be awake.

(D) Capstone tie-in

If your capstone uses a scheduled task, the ethics reflection must note the awake/app-open constraint and explain how you are handling it. If it uses a connector or computer use, note what permissions you granted (least privilege) and where you added approval checkpoints.

Objective 7 — Responsible AI: Privacy, ToS, IP, Bias, Ethics (Week 15) · 2 items

(A) Key ideas, plain language

The ethics arc comes down to two questions before every AI interaction: Is this content safe to share with an AI tool? And am I using the output in a way that is honest, legal, and fair?

(B) Vocabulary to lock

Billboard test — before pasting anything into a free AI tool, ask: "Would I be okay if this became publicly visible?" If no, do not paste it.
Data types never to paste: HIPAA (patient health data), FERPA (student education records), PCI (payment card numbers), employer confidential/proprietary material.
ToS basics — most consumer tools may retain inputs and use them for training unless you opt out; enterprise/paid plans typically have stronger protections.
IP / copyright — the U.S. Copyright Office has generally required meaningful human authorship for copyright protection; purely AI-generated work is in a contested, evolving status. Not legal advice — consult an attorney for commercial use.
Bias — AI models reflect biases in their training data; more data does not mean neutral; check outputs for fairness.
Academic integrity — submitting AI-generated work as your own, when original work is required, is academic dishonesty.
Skill 13 — troubleshooting — confused AI in a long conversation usually means context-window overload; fix = start a fresh conversation.

(C) Predictable mistakes → cures

❌ "Pasting client health data into a free tool is fine as long as I trust the company." → ✅ HIPAA obligations apply to you, not to the AI company's promises — never paste.
❌ "AI-generated content is automatically mine to copyright." → ✅ Contested and evolving; meaningful human authorship is currently required.
❌ "AI is neutral because it was trained on lots of data." → ✅ Training data reflects human biases — more data does not mean unbiased.
❌ "Editing an AI-generated essay makes it your own work." → ✅ Light editing does not constitute original authorship.

(D) Capstone tie-in

Your ethics reflection is Obj 7 applied to your own workflow: what data did you connect? What did you choose not to automate? Does your workflow pass the billboard test? Would you be comfortable if your connected-folder contents became publicly visible?

Objective 8 — Integration: The Capstone (Week 16) · 3 items

(A) Key ideas, plain language

Obj 8 is not new content — it is the synthesis. The Final will test it with scenario items: given a described workflow, identify the error, name the fix, and apply the right principle from Obj 1–7.

(B) The four capstone elements (review)

Design — what problem does it solve? which tools/features does it use?
Build — it must actually run; document the steps.
Verify — find and document at least one error; explain the fix.
Reflect — ethics reflection: data connected, what not to automate, safe-use posture.

(C) Predictable mistakes → cures

❌ "One error means the tool doesn't work — switch to something else." → ✅ One error is normal; the correct response is to document it, tighten the constraint, and verify subsequent outputs.
❌ "Automations never need review after the first run." → ✅ Automated outputs inherit AI's failure modes; review regularly.
❌ "Granting broad permissions makes the capstone more robust." → ✅ Least privilege — connect only what the task actually needs.

Fresh Practice (All New Scenarios — Vetted Answers)

None of these are live Final items. New scenarios, new wording. Cover the answers, work each one, then check. Extra weight on Obj 5–8 (the heavier back half).

Obj 1 Practice

Worked example. A student's AI assistant confidently explains that a law passed last month affects her assignment. She submits the paper and loses points because the law does not exist.
- (a) What hallucination shape is this? (b) What should she have done first?
Answer. (a) A fabricated fact (the AI generated a plausible-sounding but nonexistent law). (b) She should have looked up the law in an official government database (congress.gov or the relevant state's legislative site) before trusting the claim.

Self-check (Obj 1).
1. What is a context window? → The total text the model can hold at once; when exceeded, earlier content drops out.
2. True/False: Asking the AI to "be accurate" prevents hallucination. → False. The model generates plausible text regardless of instructions to be accurate.
3. How does a search engine differ from an AI chatbot? → A search engine indexes and retrieves real documents; an LLM generates text based on statistical patterns.
4. What does the Turing test actually test? → Whether a machine's text is indistinguishable from a human's in conversation — not intelligence, consciousness, or accuracy.

Obj 2 Practice

Worked example. A student prompt: "Summarize this reading."
- (a) What components is this prompt missing? (b) Write a stronger version with at least four components added.
Answer. (a) Missing: Role, Audience, Constraints, Format (and possibly Goal is underspecified). (b) Example: "You are an academic tutor. Summarize the following reading in 150 words or fewer, written for a second-year college student with no background in economics. Focus on the three main arguments. Use plain language and numbered paragraphs." (Role = tutor; Audience = second-year student; Constraints = 150 words, three arguments; Format = numbered paragraphs.)

Self-check (Obj 2).
1. What is meta-prompting? → Asking the AI to help generate or improve a prompt.
2. A student gives the AI two writing samples and then asks it to write in that same style. What technique is this? → Few-shot prompting.
3. What is sycophancy and how do you counter it? → The AI praises or agrees with you regardless of quality; counter by explicitly asking for critique or asking what the weaknesses are.
4. Can a simulated historical figure's "quotes" be cited as real? → No — they are generated text, never the person's actual words.

Obj 3 Practice

Worked example. A student wants to create a voiceover for a YouTube video from a written script.
- (a) What type of AI tool is designed for this? (b) Name one real tool in that category.
Answer. (a) A voice/audio-generation tool. (b) ElevenLabs (text-to-speech / voice generation). Why: chatbots generate text; ElevenLabs generates realistic spoken audio from text.

Self-check (Obj 3).
1. What are the three steps in the audio-meeting workflow? → Record → Transcribe → Analyze.
2. Which tool category is Suno in? → AI music generation.
3. Can a text chatbot generate a 60-second music track? → No — that requires a dedicated audio/music-generation tool.

Obj 4 Practice

Worked example. An AI assistant tells a student: "The minimum wage was raised to $18 per hour nationally in the United States in 2024." The student plans to cite this in a policy memo.
- (a) What verification step should she take first? (b) What hallucination shape might this be?
Answer. (a) Look up the U.S. minimum wage on dol.gov (the Department of Labor's official website). (b) A fabricated statistic (or a fabricated legal fact) — the AI generated a specific, plausible-sounding number that may be entirely wrong or outdated. As of 2024–2025, the federal minimum wage was $7.25/hour (with many state-level variations) — the AI's figure should trigger immediate verification.

Self-check (Obj 4).
1. Name three hallucination shapes. → Any three of: invented citations, fabricated statistics, fake case law, wrong arithmetic, fabricated quotes.
2. Why is asking the same AI to check itself unreliable? → The model has the same biases that produced the error and cannot reliably detect its own fabrications.
3. What does the fourth verification step ("if you're not certain, say so") accomplish? → It prompts the AI to signal uncertainty rather than generating confident-sounding fabrications.

Obj 5 Practice

Worked example. A student tells her Cowork project: "Summarize my meeting notes from last week." Claude returns a summary that invents a decision not in the notes.
- (a) What went wrong? (b) What Cowork feature would help constrain this?
Answer. (a) Claude hallucinated content — generated a plausible-sounding decision that was not in the source files. (b) Add a project instruction specifying: "Summarize only what is explicitly in the connected files; never add information that is not present in the source documents."

Self-check (Obj 5).
1. What is the difference between a skill and a connector? → A skill (SKILL.md) is a local instruction set; a connector is an MCP link to an external app.
2. What does MCP stand for, and who created it? → Model Context Protocol — an open standard created by Anthropic.
3. Is a live artifact a static file? → No — it refreshes with current data from connected apps each time you open it.
4. What is a plugin? → A bundle of skills, connectors, and sub-agents in one installable package.

Obj 6 Practice

Worked example. A student sets a Cowork scheduled task for 6 a.m. and goes to sleep at midnight with her laptop closed.
- (a) Will the task run at 6 a.m.? (b) What happens when she opens her laptop at 8 a.m.?
Answer. (a) No — scheduled tasks require the computer to be awake AND the Claude desktop app to be open. A closed laptop is in sleep mode. (b) When she opens her laptop and the Claude app launches, Cowork skips the missed 6 a.m. run and re-runs the task at that point, notifying her of the result.

Self-check (Obj 6).
1. What is the absolute rule about financial transactions and AI agents? → You — not any agent — execute financial transactions, trades, and purchases. No exceptions.
2. What is prompt injection? → Malicious instructions hidden in web page content that redirect Claude in Chrome's behavior.
3. What surface does computer use control vs. Claude in Chrome? → Computer use controls native desktop apps; Chrome controls browser tabs.
4. What is least privilege? → Connecting only the permission or folder the task actually needs — no more.

Obj 7 Practice

Worked example. A hospital administrator wants to use a free AI chatbot to draft policy summaries. She plans to paste in patient discharge reports as context.
- (a) Should she? (b) What principle applies?
Answer. (a) No. (b) HIPAA prohibits sharing patient health information with free consumer AI tools. The correct path is to use an enterprise or healthcare-approved AI platform with a Business Associate Agreement (BAA), and to anonymize any patient data even then. The billboard test also catches this: she would not want patient discharge records visible to anyone.

Self-check (Obj 7).
1. What does the billboard test ask? → "Would I be okay if this content became publicly visible?"
2. Does AI copyright belong automatically to the person who wrote the prompts? → No — the U.S. Copyright Office has generally required meaningful human authorship; purely AI-generated work is in a contested, evolving status. Not legal advice.
3. What is the Skill 13 troubleshooting move for a confused, contradictory AI? → Start a fresh conversation — context-window overload is the most common cause.

Obj 8 Practice

Worked example. A student's capstone connects a Google Docs connector and generates a weekly project status report. In the first run, the report describes a task as "completed" that is still in progress.
- (a) What type of error is this? (b) What is the correct next step?
Answer. (a) A hallucination — the AI generated a status that was not in the source document (or misread a flag in the document). (b) Document the error, add a project instruction specifying exactly how to read completion status from the document structure, and verify the next two or three runs before trusting the automation.

Study Plan — A Dated Countdown (Finals Week)

Built for the Week 16 final. Adjust the exact dates to your section. The back half (Obj 5–8) is the heaviest — once your foundations are warm, spend the most time there.

When	Do this (approx. 60–90 min)
~7 days out (end of W15)	Read this guide's Obj 1–2 sections. Work the Obj 1 and Obj 2 practice. Build your one-page vocabulary list: next-token prediction, context window, hallucination, sycophancy; all nine structured-prompt components; zero/one/few-shot definitions.
~6 days out	Read Obj 3 (modalities + tool landscape) and Obj 4 (verification workflow). Work both practice sets. Say the record→transcribe→analyze steps out loud. Recite the four verification steps from memory.
~5 days out	Read Obj 5 (the Cowork vocabulary table). Work the practice. Say the four-way skill/connector/artifact/plugin distinction out loud until it is automatic. Know: a skill is SKILL.md; a connector is MCP; an artifact refreshes; a plugin is a bundle.
~4 days out	Read Obj 6 carefully — the scheduled-task constraint (awake + app open), dispatch, computer use vs. Chrome vs. Excel, the money rule, prompt injection. Work all practice items.
~3 days out	Read Obj 7 (billboard test, data types, IP/copyright with "not legal advice" caveat, bias) and Obj 8 (capstone synthesis). Work both practice sets. Then run the Exam-Prep Tutorial (N) in an approved chatbot — it diagnoses your weak spots across all 8 objectives. Submit the share link.
~2 days out	Take the Practice Final (O) under timed, exam conditions — 25 items, treat the first attempt as real. Score it; list every missed concept by objective.
~1 day out	Re-teach only the topics you missed on the Practice Final (use this guide's mistake-cures and the relevant week's materials). Finalize and document your Capstone (design, build, verify, reflect). Sleep.
Exam day	Skim your one-page vocabulary list. Arrive rested. For each item: name the concept or principle before reading the options. For Cowork items: lock in the exact definition before choosing.

Two paired tools — use both:
- Exam-Prep Tutorial (N) — adaptive AI tutor that diagnoses weak spots across all 8 objectives and drills them; ends with a Completion Summary to submit. Best for active recall and fixing gaps.
- Practice Final (O) — 25 fresh items in the same format and blueprint as the live exam; best for pacing and a final readiness check.

(This guide points to both on purpose — it does not duplicate them.)

How the Final Is Graded + Test Strategy

How it's graded.
- 100 points across 25 items (4 each); a mix of concept, scenario, matching, and true/false items.
- The Final is 25% of your course grade — the single largest assessment. It replaces Week 16's quiz, discussion, assignment, and Studio.
- AI is not permitted on the Final.
- Coverage: Obj 1 = 3 · Obj 2 = 5 · Obj 3 = 2 · Obj 4 = 2 · Obj 5 = 4 · Obj 6 = 4 · Obj 7 = 2 · Obj 8 = 3. Back half (Obj 5–8) carries 13 of 25 items.

Honest test-taking strategies for this material.
1. Name the concept before you read the options. Every item maps to one of the eight objectives. Read the stem, say the objective out loud ("this is about the scheduled-task constraint" or "this is about connector vs. skill"), then find the option that matches.
2. For Cowork items, use the exact definitions. Skill = SKILL.md / Connector = MCP link / Artifact refreshes / Plugin = bundle / Agent = multi-step actions / Scheduled task requires awake computer + open app. If a distractor contradicts those exact facts, it is wrong.
3. For the prompting-fix item, identify what component is missing in the weak prompt (usually Role, Goal, Audience, or Constraints), then find the option that adds it.
4. For privacy items, apply the billboard test first, then check against the data categories — HIPAA, FERPA, PCI, proprietary.
5. For true/false items, look for the absolutist claim — "all citations are real," "confident tone = accurate," "scheduled tasks always run on time," "computer use and a connector are the same thing." If you have learned the exception or the correct behavior, the absolutist claim is False.
6. For matching items, assign the ones you are most confident about first, then use process of elimination for the harder pairings.
7. Read each item twice. Answer the question actually asked, not the one you expected.

Canvas Placement Block

canvas_object   = Page
title           = "Final Exam Study Guide — Weeks 1–15 (Objectives 1–8)"
module          = "Week 16 — Final Review & Exam (+ Capstone)"
grading_type    = not_graded
available_from  = 2026-12-07      # posts before the Week 16 final exam window opens
published       = true
provenance      = "~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com"

Term-update note: each term's update regenerates fresh practice variants from the same scope — the live final is never reproduced here.

~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com