Version 2 Course Development:
Design Decisions Explained

Every v2 module ends in a lab. There are no theory-only modules — if you can't apply it, you haven't finished it. The lab is not a quiz with right/wrong answers. It's a conversation with an AI that has a specific role, specific expertise, and specific instructions to challenge you until you've produced something worth keeping.

Three lab types create three different kinds of doing:

Portfolio artifacts are real outputs. Each lab has a named artifact with a specific description. A debate lab produces a written defense across multiple scenarios. A skill lab produces a completed framework applied to a live case. A build lab produces a document the learner can deploy, ship, or show. The artifact description is included in the completion signal — not just "lab finished" but "what was built."

Completion is measured by engagement depth, not time. The exchange thresholds (4–6) are calibrated so that a learner who gives shallow answers hits friction — the AI requires specificity before moving forward. A learner who engages seriously completes faster. There's no clock, and there's no minimum word count. Quality pressure comes from the AI's role, not from the system.

V2 treats responsible AI use as a structural requirement, not a topic. It appears in four places in every module — in the lesson content, in the lab guardrail, in the lab system prompt, and in the offline fallbacks. It can't be skipped because it's embedded in the mechanics of the lab itself.

The ACADEMY_GUARDRAIL is defined once per module and applied in every API call. It names the primary governance standard for that module and restricts the AI to staying on topic. The guardrail isn't a disclaimer — it's operational. It shapes every response the lab AI gives.

Governance standards are assigned, not optional. Each of the 8 modules in a course is assigned a primary governance standard from the six frameworks the platform tracks: EU AI Act, NIST AI RMF, UNESCO AI Recommendation, O*NET, AI4K12, and CSTA. That standard must be named in the lesson, applied in the lab, and cited in the COVERAGE/EVIDENCE matrix. If a module claims a standard as "primary," evidence must exist — a specific section, a specific quote.

The COVERAGE/EVIDENCE matrix holds the platform accountable. Every course has a per-standard, per-component coverage rating. Every rating above "touched" requires an evidence citation: module number, section name, and a quoted excerpt from the actual content. This is not a marketing claim — it's an auditable record of what the platform actually teaches.

Students learn to govern AI, not just use it. The debate labs specifically put governance pressure at the center of the lab. In Module 7 of the Agents course, the AI plays the role of a regulatory examiner applying EU AI Act Articles 9, 14, and 22 to deployment proposals the learner defends. The learner isn't graded on whether they get the right answer — they're evaluated on whether they can engage with governance pressure at all.

The AI model is conservative by design. V2 uses claude-haiku-4-5-20251001 — a fast, capable model appropriate for educational conversation at scale. It's not the most powerful model available, and that's intentional. Lab AI is an interlocutor, not an authority. Students should be forming their own positions, not deferring to the model's.

The platform currently serves two distinct audiences across its two versions, with different design priorities for each. V1 and V2 are not competitors — they are different points on a developmental arc.

V1 is the entry point. Foundation courses are designed for general adult and youth learners encountering AI concepts for the first time. Content uses accessible language, story-driven scenarios with named characters, short quiz questions, and manageable lab exchanges. The hub chrome provides consistent reading controls — font size, contrast, dark mode — because the learner's comfort with the interface matters as much as the content. V1 doesn't assume technical background.

V2 is for practitioners. V2 courses assume learners who are already using AI — developers, analysts, educators, professionals making real decisions about AI deployment. Scenarios don't use fictional characters because practitioners don't need narrative scaffolding; they need to see themselves in the situation directly. Labs use real framework names (NIST MAP, EU AI Act Art.13) because the learner will encounter those names in real work.

Age-appropriate isn't only about vocabulary — it's about cognitive load and stakes. For younger or earlier-stage learners, the platform manages load by breaking content into small, sequenced chunks with immediate feedback (quiz, lab, test). For advanced learners, the platform creates productive difficulty — the lab AI won't accept vague answers, won't skip to the next question, and will push back on positions that aren't supported with evidence. That pressure is appropriate for the level.

Responsible AI content is calibrated by audience. A v1 module on AI ethics asks: should this AI be used here at all? A v2 module on AI governance asks: who owns this component, what's the failure response plan, and which EU AI Act article applies? The underlying values are the same — human oversight, fairness, accountability — but the specificity and stakes differ. Younger learners build intuitions; practitioners build compliance-ready frameworks.

Cross-level design principles that apply to both:

• No account required, no paywall, no tracking — both versions are free and open
• Labs are completable offline — the fallback system fires after one retry, so a lab never stalls because the API is down
• Content avoids AI hype in both directions — neither "AI will solve everything" nor "AI will end everything"; every module grounds the learner in real decisions with real tradeoffs
• Governance is taught as a skill, not a warning — students learn to apply frameworks, not just memorize them

What's next: Future v2 courses may introduce explicit audience targeting — modules designed for K–12 educators, for healthcare workers, for policy staff. The standards alignment infrastructure (AI4K12 for K–12, CSTA for computing educators, O*NET for workforce) is already built. The COVERAGE matrix can track which courses serve which audiences at which depth. The architecture is ready for differentiation; it's a content decision, not an engineering one.