How re-theming works

The seed becomes the world.

The seed engine takes a learning objective and a per-kid signal profile and regenerates the surface around it. Sub-second for a full scene plus dialogue.

Seed structure

A seed has five components: theme, learning objective, reading level, difficulty curve, session length. The theme drives the surface (dinosaur dig, ocean dive, wizard's spell). The other four govern the mechanics; they are stable across re-themes.

  • Theme: world, vocabulary, characters, art style.
  • Learning objective: the actual skill being practiced (a spelling pattern, a phonics rule, a math fact family).
  • Reading level: sets prompt length and vocabulary band.
  • Difficulty curve: the rate at which the objective ramps within a session.
  • Session length: how many beats the session contains; sets pacing.

Regeneration time target

Sub-second for a full scene plus dialogue; longer for high-fidelity reward art, which is generated ahead of the moment it appears.

Theme rotation

{{TBD-themes}} The starter theme rotation is locked at launch. Until then, the production seed-engine surface uses placeholders.

Per-kid personalization signals

Stated interests

Topics, characters, and worlds the kid (or guardian) explicitly added to the profile.

Observed engagement

Which previous regenerations the kid stayed with, replayed, or abandoned, and at what point in the session.

Skip patterns

Whether the kid skips intros, hint loops, or reward animations; the seed engine bends pacing accordingly.

Session length

Recent session lengths inform whether to generate a short, medium, or long-form variant.

Reading level

Reading-level signal sets prompt length and vocabulary band; never used to gate access.

Agent runtime

Underneath the seed engine, R1 acts as the agent runtime that orchestrates generation, retrieval, and classification. R1 is governed by the same trust stack as the rest of FrameBright; the seed engine inherits that posture without re-implementing it.