Roadmap

Build order, and what is actually done

Updated as things ship. The order is deliberate — nothing widens until one child is genuinely learning better than they would at school.

Phases

Six phases

  1. Phase 0

    Foundations

    mostly done
    • Project, repo, and docs
    • Staging site (this site)
    • Graduate map + smart placement onboarding
    • Name decision and domain (open)
    • Model providers and secrets handling (open)
  2. Phase 1

    One kid, one verification graph, one reliable day

    next
    • Run /onboard/ for the pilot — bands + leased engine
    • Independent math + reading baseline in the first 14 days
    • Persist placement plan into the learner model
    • Narrow math verification graph + daily literacy production/oral
    • Daily plan, green day, four gates, hours and artifacts from session one
    • Two weeks of real use, and an honest write-up of what broke
  3. Phase 2

    Parent dashboard and records

    planned
    • Daily plan, hours, mastery movement, flags
    • Portfolio and evidence of learning
    • State compliance profile and exports
    • Independent achievement + growth panel
    • Transcript generation
  4. Phase 3

    All ten strands

    planned
    • Full mathematics graph
    • Reading, writing, science, history, logic
    • Computing and AI as a core subject
    • Money, physical training, practical skills
  5. Phase 4

    Tip of the spear

    planned
    • Frontier feed translated to reading level
    • Monthly shipped projects
    • Voice oral exams
    • Mentors and peer pods
  6. Phase 5

    Other families

    planned
    • Multi-family accounts, billing, onboarding
    • Pricing and packaging
    • Legal: terms, privacy, retention
    • Support and documentation

Risks

The things most likely to kill this

Written down so they cannot be quietly forgotten once the demo looks good.

Mastery could be fake
The one failure that invalidates everything downstream. Mitigated with delayed retrieval, produced work, and oral defense — and by refusing to ship vanity metrics.
Motivation without coercion
School substitutes compulsion for interest. Removing it without replacing it produces nothing. The bet is autonomy, real projects, and visible strength gains.
Cost per learner-hour
Four hours a day of large-model calls adds up fast. Small models for routine grading, large ones only for hard reasoning, aggressive caching, measured from day one.
Hallucination
Handled by grounding the tutor in authored curriculum content rather than letting it improvise.
Doing too much at once
The strongest temptation is the wide impressive demo. A narrow slice that genuinely teaches one child is worth more, so the order above will not be skipped.