How to take this check
- Do it in one sitting. 45–60 minutes is enough.
- For the multiple-choice and short-answer sections, close every AI tool and tab. This checks your internalized model, not the model’s.
- For each multiple-choice item, pick an answer before you reveal the explanation. Guessing and then reading the answer is not the same as knowing.
- For short-answer items, write your response on paper or in a text file before you reveal the model answer. Compare honestly.
- For the applied section, open your workstation-posture.md and my-first-loop.md. You are grading your own posture against the posture you actually have.
- If you miss a question, the feedback names the lesson(s) to revisit.
Multiple choice CORE
Six questions. Concept recall and diagnosis. One point each.
Show explanation
Answer: B. The four axes are the ones you route tasks along for the rest of the course: capability (how smart), cost (how much), privacy (where data goes), reliability (will it work when you need it). Review Lesson 2.1 Block 2 if missed.
Show explanation
Answer: B. A runner downloads weights and runs inference locally; it does not train, forward, or compress. (A) confuses training with inference. (C) describes a cloud proxy. (D) describes quantization, which is an input to the runner, not the runner itself. Review Lesson 2.3 Block 1 if missed.
Show explanation
Answer: D. The four rules are: never commit; never paste into chat / email / screenshots; store in .env or a secrets manager; rotate when in doubt. Password-rotation cadence is a different topic (account security) and is not the key-hygiene rule. Review Lesson 2.4 Block 2 if missed.
Show explanation
Answer: B. 3 tok/s is below the usability floor; the fix is to size down the model or route the task to cloud. (A) normalizes a broken workflow. (C) is a hardware solution for a software decision. (D) throws out the infrastructure when the fix is a smaller model. Review Lesson 2.3 Block 5 if missed.
Show explanation
Answer: B. The Cowork tab is the autonomous mode — you hand the agent a task and it works on it independently in a cloud environment. The Code tab is the interactive mode — the agent reads and writes files in a folder you've granted it, and you review every diff before it lands. Most of this course (Modules 4–8) lives in Cowork; Module 3 lives in Code. (A) reverses the pairing. (C) abandons the work tabs. (D) collapses a real distinction. Review Lesson 2.4 Blocks 1–2 if missed.
Show explanation
Answer: C. Drift is the slow kind — individual runs stay cheap, but the total adds up because nobody is looking. Counter-move: weekly dashboard check. (A) is rare and visible. (B) is what a cap catches. (D) is a separate billing issue, not drift. Review Lesson 2.5 Block 4 if missed.
Short answer CORE
Two questions, two to four sentences each. Up to 1.5 points each. Write your answer before revealing the model answer.
Q7. A student you are helping says, “I only use cloud — local models are worse, so why bother?” Name two specific problems with this stance and one concrete situation where local would be the right call even for a cloud-first student.
Model answer (for instructor / parent): A cloud-only posture has at least these failure modes: no offline fallback when the network drops or the provider has an outage; no privacy mode when the input is a journal, a medical note, or a confidence someone asked you to keep; no cost advantage on high-volume recurring work where a local model running a daily summary is effectively free; and no practice directing the smaller, more exacting models that make you a better prompt-writer. A concrete local-appropriate situation: a six-month summary of the student’s personal journal before a conversation with a counselor — the privacy axis dominates, and cloud is the wrong default regardless of capability.
Scoring: Full credit (1.5) if the answer names at least two of {no offline fallback, no privacy mode, no high-volume cost advantage, no directing practice with smaller models, no grounding in what local inference actually does} AND gives one concrete local-appropriate situation. Partial credit (0.75) for naming problems without a concrete situation. No credit for “local is free” alone — that is the beginner framing Lesson 2.1 corrects. Review Lesson 2.1 if missed.
Q8. (Optional advanced.) A classmate sends you a Slack screenshot showing their recent Claude Code CLI session (the optional advanced terminal version). You notice their API key visible in one line of terminal output in the screenshot. What should they do, in what order, and why? If you skipped Lesson 2.4's optional advanced section, skip this question.
Model answer (for instructor / parent):
- Rotate the key first. In the provider console, delete the exposed key and generate a new one. This forecloses the entire category of problem — once the key no longer works, it does not matter who saw the screenshot or where it ends up.
- Update local .env. Paste the new key into the .env file in the affected project(s) and verify the tool reconnects.
- Clean up the exposure. Ask recipients of the Slack message to delete it, and check whether the screenshot was re-shared anywhere. This step is cleanup, not protection — the key is already dead by step 1.
Reasoning. Any exposed key is a compromised key; rotation is free and instant. The order matters: if you do cleanup first, the key is live for as long as it takes you to rotate, which is exactly the window an attacker needs.
Scoring: Full credit (1.5) requires the correct order — rotate first, clean up second — with reasoning. Partial credit (0.75) for rotation without order, or order without reasoning. No credit for “change their password” (wrong control plane — API keys and account passwords are independent) or “do nothing, it is a friend” (rule 4 violation). Review Lesson 2.4 Block 2 and the “rotate when in doubt” callout if missed.
Applied CORE
Two questions, half a page each. Open-workstation — keep your posture document and loop design in front of you. Up to 3 points each.
Below are five tasks you might run in a typical week. For each task: name the dominant axis (capability / cost / privacy / reliability), decide local or cloud, and write one sentence defending the choice against a student with the opposite answer.
- Draft a college-application essay using your autobiography notes.
- Summarize your personal journal entries from the past six months for a conversation with a counselor.
- Generate a daily 7am weather-plus-calendar briefing for the next 60 days.
- Debug a React bug in a small side project’s codebase.
- Work on a short story while traveling in a national park with unreliable cell service.
Show model answer
Model answer (for instructor / parent). The routing below is one defensible path, not the only one. What matters is that each task is routed with a named axis and a coherent one-sentence defense.
- Cloud. Capability-dominant. “A college essay needs the strongest writing partner I have access to; the content is not sensitive in a way that excludes cloud; one-off runs make cost a non-issue.”
- Local. Privacy-dominant. “Journal entries are the textbook always-local category — even a cloud provider I trust is the wrong default when the input is this personal, and an 8 B local model is more than enough to summarize text.”
- Cloud with a cap, or local. Cost / reliability split. Cloud-with-cap defense: “A daily 7am briefing at even a few cents a run is still pennies a month under my cap.” Local defense: “A scheduled recurring task is exactly the shape where a free, always-available local model beats an API bill that grows silently.” Either is credit-worthy if the axis is named.
- Cloud. Capability-dominant. “Debugging React benefits from a frontier model’s deeper familiarity with framework-specific patterns; the code is a personal project so privacy is not the gating axis.”
- Local. Reliability-dominant. “In a national park I do not trust cell service; the only AI that is reliably there is the one running on my laptop.”
Scoring. Full credit (3) if all five tasks are routed sensibly with an explicit axis AND a coherent one-sentence defense. Full credit does NOT require matching this answer key — defensible alternatives count as long as the axis is named and the defense holds. 2 points if four of five are routed well. 1 point if three are. No credit for “I’d use my default for all of them” — the per-task routing muscle is the whole point of Module 2. Review Lesson 2.1 and redo the Local vs. cloud decision tree worksheet if weak.
Open your own workstation-posture.md. In your own words, identify:
- The single line in your posture you believe is most likely to fail under pressure.
- A specific situation — not hypothetical, plausible — that would cause it to fail.
- A revised version of that line that would still be true under that pressure.
Finish with one sentence on whether you will actually update your posture document after this check, and if so, what you will change.
Show model answer
Scoring. There is no single right answer. Full credit (3) if the student (i) names a specific line, not a vague “my whole posture”; (ii) describes a realistic failure scenario that is not trivial; (iii) offers a revised line that is a genuine tightening — not just a softer version of the original; and (iv) commits to a concrete edit. 1.5 points for self-reflection that does not propose a revision. No credit for “all my lines will hold” — the question is designed to surface the gap; asserting no gap is a sign of shallow engagement.
Example of a strong answer. “My line ‘I stop and audit before raising my cost cap’ will fail the first time a deadline is inside 48 hours, because the audit takes an hour I will not have. Revised: ‘Below $50 over cap, I raise once and log why in my posture file; above $50, I always audit first.’ I will edit workstation-posture.md tonight.”
Example of a weak answer. “My posture is pretty solid, I think I’ll probably keep everything the same.” (No gap named; send student back to Lesson 2.5 reflection prompt.)
Remediation. The posture is the artifact that will persist across the rest of the course. A weakness here is worth 20 more minutes now rather than carrying forward. Re-read Lesson 2.5 and rewrite one line before moving on.
Parent / instructor scoring summary
Total: 15 points across 10 questions.
- Multiple choice (Q1–Q6): 1 point each — 6 points.
- Short answer (Q7–Q8): up to 1.5 points each — 3 points.
- Applied (Q9–Q10): up to 3 points each — 6 points.
Passing bar: 11.5 of 15 or better, with at least one applied question at full credit. A miss in the applied section sends the student back to Lesson 2.5 before Module 3.
Weighting suggestions for parents issuing credit:
- Multiple choice (Q1–6): 40% of Module 2 score.
- Short answer (Q7–8): 20%.
- Applied (Q9–10): 40%. Q9 and Q10 are the load-bearing items — they demonstrate applied judgment, not recall.
Evidence to file in the student’s credit portfolio for Module 2:
- This completed check (all ten answers written out).
- The frozen workstation-posture-v1.md in the capstone folder, matching the frozen my-first-loop-v1.md from Module 1.
- The updated my-first-loop.md with all the Module-2 lines filled in (model decision; local-runner status; Claude desktop app installed and three tabs verified; default tab for autonomous work; default tab for directed file edits; editor; optional advanced API-key status if applicable).
- At least one completed per-lesson quiz and reflection from Lessons 2.1–2.5.
- A short (2–3 sentence) instructor note on Q9 and Q10 — which posture line the student revised in Q10, and what change they made.
Remediation if missed: Q1–Q6 point to a specific lesson block in the answer. Q7 → re-read Lesson 2.1 and redo the Local vs. cloud decision tree worksheet. Q8 → re-read Lesson 2.4 Block 2 and the “rotate when in doubt” callout. Q9 → the per-task routing muscle is what Module 2 is for — redo the 2.1 worksheet and revisit. Q10 → re-run the Lesson 2.5 reflection prompt; the posture must survive the rest of the course.
If the student passes at 11.5 / 15 or above with at least one applied question at full credit, Module 2 is complete and Module 3 can begin. Below that bar, target remediation to the specific lesson(s) listed above before moving on.
Next up
Module 3 — AI coding partners.
With the desktop app installed and the posture written, the next module turns the Code tab into a real collaborator: directing it on a real codebase, reading every diff before accepting, catching the moves it does badly.
Module 3 opens when the Module 2 portfolio is complete — this check, the frozen posture, and the updated loop file.