Human Original
Written and structured by a human creator. AI did not generate the instructional content.
Trust
The source should be closer than the share button.
Every Arc tells you who made it, where its claims came from, what AI changed, and when a human last reviewed it.
Human Curated · AI Adapted
A human owns the Arc and sources. GPT-5.6 may create a validated alternate explanation for one learner.
AI Generated from Verified Sources
AI proposed the content from an approved source set; a human reviewed it before publication.
Receipts
Claim → exact chunk → original source.
Every factual generated card must reference allowed source chunk IDs. The server rejects unknown IDs, missing evidence, stale content versions, and source/chunk mismatches before anything reaches the learner.
“Tilt changes direct sunlight.”→nasa-seasons-tilt→NASA Space Place
AI feedback is bounded.
Teach-back evaluation checks approved concepts only. It does not score accent, grammar, elegance, personality, intelligence, or motivation.
Failure is visible.
When GPT-5.6 is unavailable or validation fails, Skrolz says so and shows a human-approved standard explanation.
Minors are private.
Ages 16–17 are undiscoverable by default, analytics-off, and cannot publish creator content.
Data stays lean.
Raw learning text is excluded from analytics and expires by default. OpenAI calls are server-side with storage disabled.