Capstone Troubleshooting: Six Stuck Points
Six things that can go wrong.
Test drive weeks aren't always clean. These are six issues you might hit in the first month after the M10 Capstone. Each has a specific, repeatable fix; each takes under 10 minutes. Bookmark this page: you'll come back when something stops working.
Stuck point 1: your scheduled task didn't fire.
Most common Day-1 failure. The fix list, in order:
Fix recipe: task not firing
1. Are you on Pro? Free accounts save scheduled tasks but don't fire them. Check your billing page. 2. AM/PM check. Sunday at 6:00:was that 6 AM or 6 PM? The toggle catches more people than any other single thing. 3. Connector check. If the task needs your calendar (Sunday Reset, Pre-Exam Watch), is the calendar connector actually ON in Project settings? Connectors silently turn off after a re-auth. 4. Project link. Did you point the task at the Coach Project? If it's pointed at "no project," it can't read your context files. 5. App version. Old desktop versions don't have scheduled tasks at all; update Claude. 6. Still broken? Open the M9.1 Hello-Scheduled-Task plumbing test. If that doesn't fire either, it's an account-level issue, not your task: open Anthropic's help docs.
Stuck point 2: your Coach went stale.
The Coach was great in week 1, then by week 4 the output feels generic, references last term's classes, or forgets your Voice Profile. M7.4 named this "slow rot." The fix is the M7.4 maintenance routine, slightly tightened.
Fix recipe: stale Coach
The 5-minute Sunday refresh: open your Coach Project once a week (tack it onto your Sunday Reset session), and run the 30-second update rule from M7.4: (a) Add any new syllabus / new class / new prof name; (b) Remove any class that ended; (c) Refresh the standing brief if anything has shifted (new major, dropped a class, voice changed). End-of-semester gets the bigger move from M7.4: rename the Project to "[term]-archived" and start a new one for the new term. Do not just keep accumulating; that's the slow rot.
Stuck point 3: your Project got too crowded.
You started with 4 staple files in your Coach. By midterms you've added screenshots of professor emails, drafts of in-progress papers, that one PDF from a club meeting, and a half-finished personal statement. The Coach is reading all of it on every chat and the output is getting weirder. M7.4 called this "knowledge bloat."
Fix recipe: crowded Project
The trim rule: a Coach Project should have at most 8 context files at any moment, and ideally stays at the four staples + active syllabi (so 4 + however many classes). Anything else is conversational scrap and shouldn't live in knowledge: it should live inside the chat where it was relevant. Trim move: open Project context, delete every file that isn't (a) one of the four staples, (b) an active-term syllabus, or (c) a doc that gets referenced more than once a week. If a file's been there for 6 weeks and you haven't pointed at it, it's bloat.
Stuck point 4: your prompts got generic.
Sunday Reset starts producing output that could've been written for any student anywhere: generic week structure, no real connection to your actual classes, no flagging of your specific carry-over loops. Same thing happens with the Coach + your saved prompts: by week 5 the output reads like a productivity blog post.
Fix recipe: generic output
Almost always the standing brief, not the prompt. When a saved prompt runs through a tuned Coach, the Coach's standing brief is doing the work. If output is generic, the brief has lost specificity: usually because brackets never got filled, classes are out of date, or the standing orders have drifted. Move: reopen Lesson 10.2's Capstone-edition standing brief, paste it fresh, fill the brackets again with this term's actual classes / profs / quirks. Then re-run the M7.3 "what kind of Coach are you for me" tuning check. Most of the time, generic output disappears after one re-paste.
Stuck point 5: voice mode fell off.
You set up voice mode in M1.3, planned to use it on the walk-home Sunday Reset edit, and… haven't touched it in three weeks.
Fix recipe: voice atrophied
Pick one move and only one. A natural starting move is the walk-home Sunday Reset edit:it's tied to a fixed weekly ritual and uses voice for editing, not generating. Set yourself a single trigger ("when I walk out of the building Sunday afternoon, hit voice and read me my Sunday Reset"), use it once, see if it sticks. If voice doesn't fit your week shape, that's also a fine answer: running your dashboard 100% in text works just as well. You don't have to use every feature.
Stuck point 6: "I forgot to use it for a week."
Midterms ran long, you got sick, you went home for a long weekend, suddenly it's been 8 days since you opened your Coach. The dashboard didn't break; you just stopped touching it.
Fix recipe: week-long gap
Don't restart from scratch. Don't re-do the workbook. Don't re-tune the Coach. The dashboard is still wired correctly; you just haven't opened it. Move: open your Coach. Run Sunday Reset (manually if you missed Sunday: it doesn't have to be Sunday). Look at the output. Pick the one most-pressing item. Do that. Done: the dashboard is "in use" again. The fact that you went a week without it is information, not a failure: it tells you the dashboard's value is "useful when used," not "load-bearing for your life." That's the right level of dependency.
The "ignored 3 times in a row" rule is your friend
From M9.1: if you've ignored a scheduled task's output 3 weeks in a row, pause it. From 10.5: if you've cut + re-added the same piece twice, retire it. From here: if you went a week without the dashboard, that's once:restart, don't pause.
When to ask a human, not the troubleshooting list.
Most stuck points are mechanical and the recipes above will catch them. A few aren't. Three categories should send you to a real human, not Claude:
🧑🏫 Course / professor questions
"Can I use this Coach setup on this assignment?" "Does my professor's policy allow Edit-Don't-Write?" "Did I disclose AI use correctly?":these go to the professor or your school's writing center / academic integrity office, not Claude. M4.2's email templates and M0.3's cheating talk script are how you start that conversation.
💸 Account / billing problems
"My Pro renewal failed." "I got double-billed." "Scheduled tasks aren't firing and the Hello-test fails too.":these go to Anthropic's help docs and support, not the troubleshooting list above. The dashboard is downstream of a working account.
🤝 People-level problems
"My group is impossible to work with." "My roommate is impossible." "I'm worried about my friend.":these are not dashboard problems and not Claude problems. Talk to the actual people involved, your RA, your school counselor, your family. The line that ran through the whole course (M8.4 / M8.5 / M8.6) holds: AI handles the paperwork around your life, humans handle the relationships in it.
Honest Work Code · Rule 1, applied to maintenance
You stay the author. A dashboard that breaks is supposed to surface the problem to you so you can fix or cut it: not quietly degrade until you don't notice. The six recipes above all have one thing in common: they ask you to look at what's running, not to add more on top. Maintenance is a large part of the discipline of having a dashboard.
Back to the curriculum
This page lives in the bonuses library. Bookmark it; come back when something on your dashboard stops working.
Back to the curriculum →