Why Meta powers this pipeline
Every lesson pack names the Meta model that drafted it. Nothing about that is hidden, and nothing about it is hype.
1 · The packs say it on their face
All 33 library packs record Meta Model API / muse-spark-1.3 in their generation metadata, shown under every pack in the library. That metadata is hash-bound: edit one character and the checksum changes, voiding the review.
2 · Creation runs on the Muse Spark family
New drafts are created with Baseten-hosted Muse Spark Open (meta-models/muse-spark-open-1026); each draft gets one optional repair via Meta Model API with reasoning on. Every request is inspectable before sending — exact instructions, source sentences, provider, model, output limit — and needs per-request consent. Keys stay in tab memory; outcomes land in an exportable usage ledger.
3 · New: a Meta-vision author path
Photograph a printed notice you have the right to use, and Meta vision transcribes the visible English into the source form for your review — inspect the exact payload, consent per request, outcome in the usage ledger; transcription only, at least 100 legible characters or it is rejected. The transcribed reading then walks the same road as typed text: selection, exercises, review, approval. No step is skipped because the words came from a photo.
4 · Tutor defaults to Meta
The main-app Tutor defaults to Meta Model API / muse-spark-1.3; other providers are opt-in BYOK with the same per-request consent and quote checks.
5 · Why not another model?
An honest answer, not a superiority claim: the grounding contract (verbatim phrases, exact cloze reconstruction, no auto-graded writing), the 17-criterion review gate and the immutable-pack pipeline were built and validated against Muse's JSON discipline. Swapping providers means re-reviewing packs under new hashes — the pipeline makes that possible, but we have not done it, and we claim no comparative advantage. No learning gains are measured or claimed.
AI-assisted review is not human certification. The demo uses retained material and Chrome offline emulation. Details: offline pipeline dossier.