Pilot and impact
Twelve weeks, one school, one honest comparison
The pilot is designed so that “printed activities did just as well” is a result we could publish rather than a failure we would hide.
Foundation and OUSD select one elementary school and a book distribution event. Interview caregivers and educators. Appoint owners. Agree budget, consent, and access needs.
Prepare 12 activities. Configure and test the proposed service. Review accessibility, content rights, translation, privacy, opt-out, deletion, and human handoff.
Invite 60 households. Offer two activities weekly. Where feasible, randomly assign half to guided AI and half to equivalent printed activities with equal facilitator support.
Review results, costs, incidents, and family feedback. Decide whether to improve, expand, or retain the simpler approach.
Procurement or platform setup may extend preparation beyond weeks 3–4.
Targets—not measured results
- 70% of households complete at least eight activities.
- 70% of participants with paired assessments improve by at least one point on a 0–4 rubric.
- Investigate participation gaps above 15 percentage points across language or access mode where privacy permits.
- Zero critical failures in the prelaunch safety test suite.
Nothing has been measured. These are the numbers we would aim at and report against, whichever way they come out.
The 0–4 rubric
One point each for:
- a specific question
- cited evidence
- a detected error
- a useful revision
Scored on paired samples of the child's own work, by an educator — never by the Coach.
Scenario calculator
Planning scenarios only. No budget is supplied, and nothing here is a measured result.
Expected completing households
42
Cost per completing household
—
Enter a total pilot cost to see this scenario.
Cost categories to budget
- Staff time
- Content review
- Translation
- Accessibility testing
- Platform licenses
- Messaging
- Printing
- Family participation support
No figures are attached: a fabricated budget would be worse than a blank one.
Efficiency choices
- Reuse approved content instead of generating it per family.
- Use rules, not a model, for routine navigation.
- Keep prompts and responses short.
- Limit unnecessary model calls.
- Avoid generated media.
- Finish activities offline, away from the screen.
These choices reduce cost and compute by design. We make no claim of measured emissions reductions.
Beyond the pilot, if the answer is yes
The pilot is deliberately small, but the shape of the work is what makes a second site cheaper than the first. One reviewed activity library, one support queue, and one set of participation records serve EAT, LEARN, and PLAY together. Adding a second school means adding families and facilitators to something that already exists, rather than commissioning a second build. Adding a second program means adding passages and activities to the same reviewed library, under the same review step.
What would have to be true before expanding
- Educators say the paired work samples show real gains, and printed activities did not do just as well on their own.
- Participation held up across languages and access modes, not only among families with the easiest access.
- Facilitator time per family was low enough that a second site does not need a second full-time person.
- The safety test suite passed clean, and no incident required the pause procedure.
- The real running cost per completing household is known, not estimated.
No expansion is proposed or funded here, and no reach, cost, or savings figure is claimed. This describes what would have to be shown first.
The decision for the foundation
Start with one school, one book event, and a clear learning question.
The first decision is whether this is useful enough to explore with families and program staff. A small pilot would test both learning value and the effort required to operate the service.
Nothing here is funded, approved, or partnered. The proposal asks for one thing: permission to prepare a small, honest test at one school, alongside one book distribution event, with a learning question specific enough that the answer could be no. If printed activities with the same facilitator support do just as well, that is a real and useful finding, and the foundation should keep the simpler approach.
Proposed next steps
- 1Identify a foundation program sponsor.
- 2Select a literacy partner and potential school event.
- 3Review one activity with caregivers.
- 4Validate technology and operating costs.
- 5Decide whether to authorize pilot preparation.
No funding, partnership, or approval has been secured.