How to take this check
- Do it in one sitting.
- 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 frozen /capstone/automation-register-v1.md, the retirement-ritual worksheet, and your prompt templates. Q9 and Q10 are open-workstation on purpose.
- 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. Review Lesson 6.1 Block 2 if missed. (A) is the Module 5 shape with a wrong fifth step. (C) is the Module 4 research-agent shape. (D) looks plausible but is a loose product-development paraphrase, not the Module 6 structure. The Module 6 shape specifically names the parts that exist only because work is unattended.
Show explanation
Answer: B. Review Module 6 README and Lesson 6.1 Block 3 if missed. The rule is uniform; “it’s routine” and “the family is okay with it” are precisely the carve-outs the lesson warns against. (D) is false — calendar actions that reach other people are explicitly in scope.
Show explanation
Answer: B. Review Lesson 6.2 Block 5 if missed. The date-as-idempotency-key trap is time-zone drift; the recipe’s TZ=<canonical> fix pins the filename to your home zone so travel is harmless. Students who did not take the fix will produce overwrites or day-shifted files.
Show explanation
Answer: C. Review Lesson 6.3 Block 5 and Lesson 6.5 Block 1 if missed. An always-empty health footer across time is the cheapest visible signal of a lazy refresh. The fix is to restate “search all sources every run, report what you searched” in the prompt and verify the footer changes run-to-run.
Show explanation
Answer: B. Review Lesson 6.4 Block 5 if missed. The three register columns decode three distinct failure modes. In this pattern: scheduler fine, no condition met, whatever fires the student reads. The right question is whether the real-world event that retires the watcher has already happened — the retirement-trigger column, not the fire count, is the decider.
Show explanation
Answer: D. Review Lesson 6.5 Block 4 if missed. Age alone does not retire a task. The five real signals are named event, stale last-success, silent drift, ceiling breach, and human-attention bankruptcy. A twelve-month-old automation that still earns its keep and has a future retirement trigger is a success, not an obsolete artifact.
Short answer CORE
Two questions. Three to four sentences each. Up to 1.5 points each.
Q7. A classmate says: “I set up a scheduled task that auto-replies ‘got it, thanks!’ to any routine newsletter that asks for confirmation. It’s not a real relational send, so the Module 6 rule shouldn’t apply.” Name three specific problems with this stance, referencing Module 6 concepts.
Rubric. Full credit (1.5) names at least three of the following:
- The audience-only-you rule is uniform and does not have category carve-outs in this module.
- “Routine” is the same carve-out the drafts-not-sends rule in Module 5 already declined; carve-outs metastasize.
- Newsletters that ask for confirmation are outbound to a third-party system that treats the reply as consent — a send, not routing.
- The failure mode is silent, because the student will not notice a wrong auto-reply that the other side processed.
- An audience-only-you version of this automation exists — a watcher that flags newsletters needing confirmation so the student can click-reply themselves — and is the correct shape.
Partial credit (0.75) for two reasons with the uniform-rule framing intact. No credit for “you just shouldn’t” without specific failure modes. Remediation: re-read Module 6 README and Lesson 6.1 Block 3.
Q8. A student reports that their weekly report’s Said-I-would, didn’t section has been empty for three consecutive weeks, while their Shipped this week section is populated. What is the specific failure mode, why is it suspicious, and how does Module 6 recommend fixing it?
Rubric. Full credit (1.5):
- Names the input-scope failure — the agent is summarizing shipped work from git but has no source for carried commitments (probably the drafts folder, sent folder, or open-thread inputs the prompt named are not scoped correctly or were not included).
- Explains why always-empty is a signal — real weeks always have carried items; the section being empty across three weeks means the agent is not seeing them, not that they do not exist.
- The fix is to audit the weekly report’s input scope, add explicit sources for carried commitments (drafts folder count, threads with open replies, capstone-file TODOs), and re-run.
Partial credit (0.75) for naming the failure mode without the fix. No credit for “the student is efficient and finishes everything.” Remediation: re-read Lesson 6.3 Block 2 (the five-section weekly report) and re-run the Lesson 6.3 activity with an explicit list of drafts/open-thread inputs in the prompt.
Applied CORE
Two questions. Half a page each. Up to 3 points each.
Scenario. You built a watcher in Lesson 6.4 called Watcher — teacher reply. Its condition-and-threshold statement was: “Fire when any email arrives in agent-access/school from the domain of Ms. Rivera’s school district.” Its quiet-run behavior writes a one-line log entry. Its response action is “open the thread within 30 minutes and draft a reply.”
Four weeks in, your register shows: last run = 90 minutes ago (healthy), last fire = 19 times in 28 days, last success = 2 weeks ago (you read and acted on the first 9 fires, and have not opened an alert since). The prompt-change log is empty.
Task. In 8–12 sentences:
- Name the specific failure mode by its Module 6 term.
- Diagnose which part of the condition-and-threshold statement is the likely cause. Be specific about the words in the statement.
- Rewrite the condition-and-threshold statement to fix it — event, qualifier, threshold, non-event — referencing the watcher-design worksheet structure.
- Name the one explicit non-event you added and explain why that non-event is the lever that reduces fires the most without increasing false-negative risk.
- State whether you would retire rather than redesign, and on what grounds.
Show model answer
Model answer (sketch).
- Failure mode: alert fatigue. 19 fires, 10 not opened, last success two weeks ago while last fire is current. The watcher is accurate enough to fire but the student has stopped reading.
- Likely cause. The qualifier “from the domain of Ms. Rivera’s school district” is domain-level scope — the watcher fires on any email from the entire district, which includes mailing lists, administrative notices, and other teachers. The fix is to tighten the qualifier to Ms. Rivera specifically, plus a thread-level qualifier.
- Rewrite:
- Event. An email arrives in agent-access/school from rivera@<district>.
- Qualifier. The email is either a reply in a thread I sent to, or a new thread whose subject names me or a course I am enrolled in.
- Threshold. Fire only on the first new Rivera message per thread per 24 hours (subsequent replies in the same thread do not re-fire).
- Non-event. Auto-acknowledgements, read receipts, district-wide announcements CC’d to the address, and mailing-list traffic do not fire.
- Non-event that moves the needle. “District-wide announcements CC’d to the address do not fire” — these are what exploded the count, and excluding them does not risk missing a real Rivera-to-me message (which goes To, not CC).
- Retire vs. redesign. Retention is at 47% (9 of 19 read and acted). If I can commit honestly to reading the narrower watcher, redesign. If I cannot commit, retire — a watcher I will not read fails the response test retroactively.
Scoring. Full credit (3): names alert fatigue; identifies the domain-level scope as the over-loose qualifier; rewrites with a specific sender, a thread qualifier, a per-thread/per-24h threshold, and at least one named non-event; defends the non-event choice by linking it to the observed false-positive pattern; makes a defensible retire-vs-redesign call. Partial credit (1.5): names fatigue and one or two of the four structural fixes. No credit: “the watcher is broken” without diagnosis.
Remediation: re-read Lesson 6.4 Block 2 (three failure modes) and Block 3 (condition-and-threshold anatomy). Re-run the Signal-vs-noise drill.
Scenario. Open your frozen /capstone/automation-register-v1.md. Pick the task in your register that you predict will be the first one you actually retire in the next six months — even if you have not retired it yet.
Task. In 8–12 sentences:
- Name the task and its current register row status.
- Identify which of the five retirement signals is most likely to fire first. Be specific: if named event, name the event. If stale last-success, name your own threshold and how close you are to it. If silent drift, describe the drift concretely. If ceiling breach, compute the current monthly cost and name the ceiling. If human-attention bankruptcy, describe the attention budget and where this task falls in it.
- Walk through the three retirement steps you will perform when the signal fires: disable (which scheduler and how), revoke (which credentials and where), archive (which folder and the file-name convention).
- Name the one credential you must specifically remember to revoke — the one that would otherwise linger as a zombie token.
- Write the one-sentence post-mortem you expect to leave in the register row when the task retires.
Show model answer
Model answer (illustrative — your specifics must come from your own register).
- Task: Watcher — State v. Patel keyword. Status: Live, last fire 9 days ago.
- Signal: Named event. The retirement trigger is “the State v. Patel case is decided or withdrawn.” The California Court of Appeal calendar lists oral argument in 7 weeks; a decision typically follows inside 90 days. I expect the named-event signal to fire within 4 months.
- Disable: launchctl unload ~/Library/LaunchAgents/ai-directed.watcher-patel.plist, then move the plist into ~/Library/LaunchAgents/retired/.
- Revoke: the task holds a scoped Perplexity API key it does not share with any other task. Delete the key on the Perplexity dashboard, confirm revocation with a 401 on curl.
- Archive: mv ~/ai-architect-academy/automation/watcher-patel/ ~/ai-architect-academy/automation/retired/watcher-patel-2026-08-14/ (substituting the actual retirement date).
- Zombie credential to name explicitly: the Perplexity API key. It is specific to this task; if I skip step 2, it sits in my credentials folder indefinitely.
- Post-mortem: “Fired 12 times over 7 months; acted on all 12; retired cleanly when the appellate decision ended the topic. Worth rebuilding if I track another case with a similar read-and-act posture.”
Scoring. Full credit (3): a specific task named from the student’s own register; the correct retirement signal identified with an explicit computation or observation; the three-step ritual spelled out accurately with the specific scheduler, credential, and folder convention; the specific token-to-revoke named; a post-mortem that is honest about what worked and what did not. Partial credit (1.5): a specific task and correct signal but vague ritual steps. No credit: a generic “I’ll retire it when it stops being useful” answer without specifics.
Remediation: re-read Lesson 6.5 Block 4 (five retirement signals) and Block 5 (revoking schedule-bound credentials). Run the retirement ritual once on a real task from your register, even if the task is not yet ready to retire — walk through the three steps mentally with your specific folders and tokens in front of you.
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 on Q9 sends the student back to Lesson 6.4 and the Signal-vs-noise drill; a miss on Q10 sends them back to Lesson 6.5 and the retirement-ritual worksheet before Module 7.
Weighting suggestions for parents issuing credit:
- Multiple choice (Q1–6): 40% of Module 6 score.
- Short answer (Q7–8): 20%.
- Applied (Q9–10): 40%. Q9 and Q10 are the load-bearing items — they demonstrate applied judgment on a real watcher diagnosis and a real retirement plan, not recall.
Evidence to file in the student’s credit portfolio for Module 6:
- This completed check (all ten answers written out).
- The frozen /capstone/automation-register-v1.md — six entries + Module 6 reflection paragraph, at least one row marked Retired <date> with a post-mortem.
- The /capstone/automation-artifacts/ folder — representative outputs from every live task (a morning brief, a weekly report, a research refresh, a watcher firing artifact + quiet-run log, and entry 5).
- The Signal-vs-noise drill summary saved from Lesson 6.4.
- The updated my-first-loop.md — current Automation Posture section reflecting the numbers the student actually saw.
- A short (2–3 sentence) instructor note on Q9 and Q10 — what failure mode the student diagnosed in Q9, and which task-plus-credential the student identified in Q10.
Transcript language. If a parent is assembling a transcript, the transcript line for Module 6 can accurately say: “Directed AI agents to run unattended on a schedule — daily briefs, weekly reports, research refreshes, and conditional watchers — with a documented automation register, per-task cost ceilings and next-review dates, and a formal retirement ritual including credential revocation.”
Remediation if missed:
- Q1: Re-read Lesson 6.1 Block 2 and the Automation Posture statement in my-first-loop.md.
- Q2: Re-read Module 6 README and Lesson 6.1 Block 3. Audience-only-you is the module’s load-bearing norm.
- Q3: Re-read Lesson 6.2 Block 5 (the TZ trap). Run the morning brief through a manual time-zone shift to feel the failure mode.
- Q4: Re-read Lesson 6.3 Block 5 (the diff contract) and Lesson 6.5 Block 1 (observability floor). Audit your research refresh’s health footer.
- Q5: Re-read Lesson 6.4 Block 5 (the three-column watcher audit).
- Q6: Re-read Lesson 6.5 Block 4 (the five retirement signals). Print them and keep them near your workstation.
- Q7: Re-read Module 6 README and Lesson 6.1 Block 3. The carve-out pattern is the trap the module exists to close.
- Q8: Re-read Lesson 6.3 Block 2 (the five-section weekly report). Re-run the weekly with an explicit drafts/open-thread input in the prompt.
- Q9: The watcher-diagnosis muscle is the load-bearing skill of Module 6. Re-do Lesson 6.4’s Signal-vs-noise drill and sharpen the condition-and-threshold statement on your own watcher.
- Q10: The retirement ritual is the load-bearing discipline of Module 6. Walk a real task through disable–revoke–archive, even if you are not yet retiring it.
If the student passes at 11.5 / 15 or above with at least one applied question at full credit, Module 6 is complete and Module 7 can begin. Below that bar, target remediation to the specific lesson(s) listed above before moving on.
Next up
Module 7 — Extending AI (skills & plugins).
The automations from Module 6 are the natural candidates to extend into custom skills. Module 7 teaches the student to build reusable skills and plugins that live inside the agent — shared prompts, shared tools, shared discipline — and to judge when a skill is worth building vs. when an off-the-shelf one already earns its keep.
Begin Module 7 once the Module 6 portfolio is complete — this check, the frozen automation register, the artifacts folder, and the Signal-vs-noise drill summary.