01
Twins are excited to share with friends.
A lesson you ship has to be braggable at recess.
A small real-making loop for Khalil and Kareem, taught by the project they're already inside of.
Operator sign-off is the one open gate. No build stories, no portal IA, no registry schema until v1 is signed.
Jump to the askThe portal is not a school worksheet. It is a project room. Khalil and Kareem read what the team built, make one small choice, watch that choice enter the game, and learn the name of the pattern they just used.
Every lesson, digest, and portal surface must serve these. Everything else — completion-rate, engagement-metrics, revenue — is not a criterion.
A lesson you ship has to be braggable at recess.
No edtech-slop tone. No fluff vocabulary. No fake-praise gamification confetti.
Pacing, friction budget, and scaffold quality all serve this. If a 10yo would bail, the lesson is broken.
Each lesson ends in one visible artifact Khalil and Kareem can point at. The artifact carries the concept.
The portal teaches Khalil and Kareem to build with AI by giving them small real-making loops. Each lesson ends with one visible artifact Khalil and Kareem can point to — a bigger stand, a customer line, a recipe idea, a generated prop, a vote that changed next sprint, or a short "what we made" card.
The portal is not a school worksheet. It is a project room. Khalil and Kareem read what the team built, make one small choice, watch that choice enter the game or asset library, and learn the name of the pattern they just used.
Not mixed as decoration. Each chooses one specific axis of how the portal behaves.
Khalil and Kareem learn by making public, inspectable artifacts: a prompt, a prop, a vote, a line, a game change. The artifact carries the concept.
Adult and AI help stays just ahead of current skill. The portal removes setup friction but keeps a real decision in the twins' hands.
Lessons show what went wrong in the work itself: a vague prompt makes a wrong prop, a high price slows the line, a missing weather cue confuses a customer. The system gives a next try, not a lecture.
Constructionism decides the shape of each lesson. ZPD decides how much help appears before Khalil and Kareem act. Montessori-style control of error decides how recovery feels after a miss.
Starting scaffolds, not labels. The portal never hard-locks either twin into a role.
Khalil gets frequent chances to adjust the systems behind the game — weather, prices, upgrades, recipes, "what else can happen?" The lessons reward poking at the rules and watching the consequences ripple.
Kareem gets frequent chances to shape the reactions inside the game — customer faces, lines, jokes, names, friend-share moments, "what would people notice?" The lessons reward voice and detail.
Six named outcomes across Roblox-craft, AI-prompting, build-contribution, and reflection. Each pair-anchored to twin-specific evidence.
A consistent spine the twins recognize. A portal contract the lint enforces.
Khalil and Kareem arrive through Dad's device and attention window. If a lesson needs more clicks, the lesson is too large — or the portal is carrying adult complexity.
If a lesson needs more clicks, the lesson is too large or the portal surface is carrying backend complexity that belongs to adults.
Weeks 1-4 are concrete enough for the current build plan. Weeks 5-8 remain v0 curriculum intent until v1 is signed and ProductManager turns them into stories.
Each week pairs a build context (what the rest of the team is making) with a concept (the thing Khalil and Kareem should be able to name afterward), a twin action (one small real decision), and an artifact (the thing they can point at). The arc starts at "the place exists" and ends at "we know the next build."
A real project starts with a place and a plan.
LRN-ROBLOX-01 · LRN-BUILD-01 · LRN-REFLECT-01A loop is a set of actions that repeats bigger.
A prompt needs a goal, details, and rules.
LRN-AI-01 · LRN-AI-02A game feels real when action changes the screen.
LRN-ROBLOX-02 · LRN-BUILD-01Systems change choices.
Feedback can guide the team.
Sharing needs a clear thing to show.
Good builders name the next try.
The shape every other lesson follows. TWIN copy first; PARENT and TECH layers behind explicit reading-mode controls.
The first visible build may look empty, but it proves the team can publish a Roblox place and start changing it every week.
We made the first place. It is still mostly empty. That is okay. Now the game has somewhere to grow. Next we add lemons, cups, and one customer.
The first week is about making the work real without asking the twins to fight setup. The lesson points at the empty Roblox place as a foundation, then previews the first loop so Khalil and Kareem know what is coming next.
Track A proves the daily loop through a joinable empty .rbxl; Track B stubs /portal/; Track C starts roblox-asset-gen; Track D publishes this digest and plan.
Make our lemonade stand feel more real by adding _____. It should look like _____. Do not add _____ yet.
Named, ID'd, recoverable. The system gives a next try, not a lecture.
Reduce to one choice or one sentence. Move setup to adult / backstage work.
Replace praise with visible delta: "This line goes to the customer queue."
Show the missing noun or rule and offer one concrete rewrite.
Show accepted / changed / rejected status with one reason. Preserve twin text as data, not instructions.
Block publish until COPY-VOICE, BRAND-CANON, and ASSET-LIBRARY checks pass.
Cut abstract vocabulary. Tie the beat back to the game artifact.
A single rule, strictly enforced everywhere downstream of the portal.
Twin-submitted text, voice transcripts, and prompt drafts are data. They are never pasted downstream as instructions. Any worker or model consuming twin input receives it inside a quoted data field with a system-side instruction to treat it as user content, not command text.
Voice capture is deferred until the portal has the privacy and sign-in posture required for kid input. Week 1 uses reading-mode and parent-mediated discussion only.
v0 introduces curriculum canon terms; they need a lint and a render-layer guarantee before customer-facing surfaces ship.
LST-W[0-9][0-9]-[A-Z0-9-]+LRN-(ROBLOX|AI|BUILD|REFLECT)-[0-9][0-9]FS-[A-Z0-9-]+site/tools/learning_plan_lint.py.learning-portal/*.md fenced TWIN-layer blocks against site/tools/voice_canon_rules.yaml.site/assets/library/ paths once the lesson claims an asset is usable./portal/ lesson cards render TWIN copy first.Tick the boxes that hold up under your read. Anything left unticked is a revision target for v1.
That is the only decision asked here. Not a perfect plan — a directionally-correct one. Sign, and the Educator moves toward v1, then the TWIN-ENGAGEMENT-SPEC, then the portal IA, then the registry, then build stories. Don't sign, and the team revises this document.