Project status, not conventions. Constraints live in CLAUDE.md, in the repo root.
Session 1 has now been delivered to both age groups. That makes the middle school track complete — MS is Session 1 and ends there. High school still has Sessions 2, 3, and 4 ahead, and none of those have been run with campers. This file is rewritten around that split: what the first delivery settled, and what is still an estimate on paper.
Recorded 2026-08-10.
Session 1 running at all means a whole column of assumptions came good. Don’t re-verify these from scratch — check them the morning of, out of habit, not out of doubt.
| Assumption | Status after Session 1 |
|---|---|
claude launches on a lab machine |
Confirmed — the session happened |
python -c "import tkinter" works |
Confirmed — campers got windows on screen |
Documents\Projects exists / can be made |
Confirmed |
Name folders (mkdir <Name> / cd <Name>) |
Confirmed — setup depended on them |
The 6-minute dice.py opening demo |
Confirmed — improvised live, it worked |
| Google Drive take-home, both age groups | Confirmed as far as Session 1 goes |
| MS Session 1 fits 85 minutes | Confirmed — ran the full 85, with a break, pacing judged right |
Copy-Item / Remove-Item save-point cycle |
Still unverified — Session 3 only |
| HS Session 4 demo pacing | Still unverified — Session 4 hasn’t run |
| Session 2 bad-prompts handout in practice | Still unverified |
The save-point line is the one that matters. Nothing in Session 1 exercised the folder copy commands, so the lab-permissions risk that could break the entire Session 3 save point system is exactly as unverified as it was before you taught anything.
The two deliveries disagree, and the disagreement is the useful part:
| Planned | Actual | |
|---|---|---|
| MS Session 1 | 85 min | 85 min. Fit, with a break. Pacing judged right. |
| HS Session 1 | 120 min | ~60 min. Half the budget. |
The MS estimate was good. The HS estimate was inverted. Session 1’s plan assumed high schoolers need about 1.4x the middle school durations for the same block sequence. They actually got through it in about 0.7x — faster in absolute terms than the younger group, not slower. The content wasn’t wrong; the assumption that older campers consume more clock was.
That matters for what’s left, because Sessions 2 and 3 were estimated the same way. But don’t apply a flat discount — the compression is in the talk and structured-exercise blocks, not the build blocks. A 75-minute build block is 75 minutes of building regardless of age. What collapsed in Session 1 was the talking.
| Session | Planned | Where the risk is |
|---|---|---|
| HS 2 | 180 min | The one to watch. ~70 min of it is talk and lab (warm-up, prompt sizing, prompt lab, the four checks). Expect that front half to run well short. |
| HS 3 | 180 min | Lower risk — 75 min of it is capstone build, which will use its time. The opening 50 min of talk may compress. |
| HS 4 | 60 min | Demos scale with class size, not with the estimate. Unaffected. |
If Session 2’s front half collapses you have an hour of dead room with campers who have pitched a capstone and can’t properly start it until Session 3. Have overflow ready before you walk in — the AI topics segment bank was written for exactly this, and any of its six segments drops into Session 2 as easily as Session 1.
Leave the MS timings alone. They are the one set of numbers in this repo that has been measured against real campers and come out right.
These are the things only you know now, and they’ll be gone in a month. Replace each line with the answer.
troubleshooting.html? Add it
there, not here. →Trimmed to what’s genuinely still ahead. Each item names what breaks if you skip it.
Walk the save-point cycle on a real lab laptop. Copy-Item -Recurse to save,
Remove-Item -Recurse then Copy-Item -Recurse back to undo, inside a <Name>
folder. Do the File Explorer version too (right-click → Copy → Paste → rename).
This is the highest-risk unverified thing left in the track. If lab permissions
block the copy commands, the entire save-point system in Session 3 breaks, and
Session 3 is where the capstone has to be finished. Session 2 also ends with campers
making a capstone-working copy, so it fails there first.
Confirm class size for HS. Session 4 has 30 minutes of demo time; at 90 seconds each that’s 20 campers with zero slack. Above ~20, cut to 60 seconds or run part of it as a gallery walk — decide before the day, not during it.
Have the overflow material ready. See the timing finding above. Pick your segments from the topic bank before the session, not from the front of a room with 40 minutes left.
Print the Session 2 handouts. Camper notes one per camper, the bad-prompts sheet one per pair (it lives inline in the lesson plan), and spec sheets one per camper plus spares.
main, folder /docs. Click through home → High School → a session’s notes
→ back, and open it on a phone.The tone of the “the AI wrote the code” talks at the end of Sessions 1 and 4. The most opinionated thing in the curriculum. MS gets two sentences, HS the full paragraph. You’ve now delivered the MS version once — if it landed, leave it alone.
Whether the MS break becomes the default. It’s currently written as “no break scheduled, take 5 at 0:48 if your group needs one.” A break was taken, the session fit, and the pacing was right — so the note arguably has it backwards. Flipping it means editing the MS timing table, so it’s your call, not a correction.
Whether MS gets a topic segment. Settled: no. MS Session 1 ran its full 85
minutes and the workload was judged about right, so there is no slack to spend. The
topic bank stays HS-only. Anything added to MS displaces
something that’s already earning its place.
So nobody “fixes” these by accident.
session-1/opencode-zen-howto.md
covers using OpenCode’s free models so campers can keep coding at home at no cost. As of
2026-08-10, every “keep building at home” line on the site links to it via a \*
footnote — middle-school.html, session-3/hs-camper-notes.md,
session-4/hs-camper-notes.md. It is still not in teacher.html, the main nav, or
any resource list — that’s now a judgment call rather than something still blocked.
Update, 2026-08-11: the free models turn out to need no account, no API key, and no
provider setup at all — they’re available the moment OpenCode is installed. That
dissolves the earlier CLI-vs-desktop-app split; there’s no connect flow to get wrong.
Confirmed working on a personal Windows 11 machine (2026-08-10) and, this update, on an
actual CompuCon lab machine (2026-08-11) — the one gating item left. The only open
question now is the usage cap: one user reports hitting 200 requests in a 5-hour
window, unconfirmed whether that’s per-model or total. The doc still intentionally
breaks the one-command rule (opencode vs claude), which is why it lives as a
footnote rather than class material — but that’s a smaller departure than before, since
it no longer breaks the no-accounts / no-API-keys rules too.local-models-howto.md — the Claude Code half is confirmed, the OpenCode half
isn’t. Added 2026-08-11, a third “keep coding at home” option — running a model on
your own computer with Ollama or LM Studio. Confirmed 2026-08-11: claude
connected directly to Ollama (the env-var method, no proxy) has been run for real on
both a Mac and a Windows machine. Still unverified: the OpenCode-with-Ollama
config, and the install/hardware specifics, all still written from docs rather than
from running them. It’s a home-only option — no CompuCon lab laptop has the hardware
for it. faq.html now links to it from the “without the cloud” question.docs/session-N/lesson-plan.md renders to a
/session-N/lesson-plan.html page, and teacher.html links all four. Anyone who
finds the site can read them. Fine if the plans aren’t secret — worth knowing if
they are.style.css.
The pandoc template that used to wrap those pages in the site theme is gone. Known
and accepted; if site-wide consistency ever matters, a Jekyll layout committed to
the repo would fix it.
This Vibe Coding Curriculum by Brian Bird, created in with AI assistance, is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.