Evidence and citations
These studies and published results shape what Foundry expects to test. They do not prove a Foundry outcome, and unlike measures are never averaged into one convenient number.
Operator reports and controlled studies are labelled separately.
Every source includes the limitation that matters most.
The quantitative evidence set behind Foundry's expectation of unusually high learning efficiency.
What it reports
The operator reports 2.6 times average annual growth on MAP assessments, with its highest-growth students reaching 6.5 times growth, inside a full-time two-hour-learning school model.
How we use it
This is an operator-reported result, not an independent Foundry result. The country, schedule, admissions, assessment window and student population differ from the Foundry pilot.
Open the original sourceWhat it reports
The operator describes a 90 percent mastery threshold, adaptive core academics before collaborative workshops, and motivation systems that combine individual goals, collective rewards and an internal currency called Alphas.
How we use it
This describes another operator's implementation. It informs questions for Foundry to test rather than proving the same implementation or result here.
Open the original sourceWhat it reports
In a lottery-based evaluation of an after-school program, students gained 0.37 standard deviations in mathematics and 0.23 in Hindi over 4.5 months. The paper describes roughly twice the control group's math progress and 2.5 times its Hindi progress.
How we use it
The intervention combined adaptive software with instructor-led small-group learning and served a different population. It does not isolate a universal multiplier for every platform or learner.
Open the original sourceWhat it reports
A cluster randomized trial across 44 schools reported 75 percent more mathematics learning than the normal annual gain, with an effect size of g = 0.22. Lower-performing students showed the largest benefit.
How we use it
ASSISTments was embedded in a year-long school and teacher-development model. Its result is evidence for timely feedback and targeted practice, not a direct forecast for Foundry.
Open the original sourceWhat it reports
In a short randomized study of 13 to 15-year-olds in Chengdu, the adaptive group showed 4.19 times the pre-to-post gain of students taught the same topics by expert classroom teachers, with Hedges' g = 0.68.
How we use it
The study lasted three days, had substantial attrition and was authored by the platform's research team. It is a striking short-run result that needs cautious interpretation and longer replication.
Open the original sourceWhat it reports
A three-year panel covering more than 200,000 students found an estimated 0.031 standard-deviation mathematics gain at about 11 minutes of Khan Academy practice per week, rising to a projected 0.085 at 30 minutes.
How we use it
This was a large observational design rather than a randomized trial, and two authors were Khan Academy employees. Usage quality and implementation affected the result.
Open the original sourceWhat it reports
The study found that greater use and following the recommended lesson sequence were associated with faster gains. Average usage corresponded with a two-percentile-point MAP gain in one district.
How we use it
The researchers described the causal evidence as encouraging but mixed and could not rule out motivation or teacher effectiveness as explanations.
Open the original sourceWhat it reports
This meta-analysis connects autonomy, competence and relatedness with more self-directed motivation, persistence and academic outcomes.
How we use it
Motivation is shaped by context and relationships. A reward system or software feature cannot create it automatically.
Open the original sourceWhat it reports
The platform defines one XP as one minute of focused learning, then adjusts awarded XP using mastery, efficiency and evidence of gaming rather than treating all screen time as equal.
How we use it
This is another platform's operating framework, not independent evidence of learning impact. Foundry uses it as a design reference and must test whether its own rules reward the intended behavior.
Open the original sourceSupporting context for mastery, motivation, screen use, assessment and the Chiang Mai setting.
What it reports
National norms provide a way to interpret both achievement and growth relative to a large reference population.
How we use it
A norm is a comparison framework, not evidence that a specific learning model caused the observed growth.
Open the original sourceWhat it reports
The evidence summary finds that mastery approaches can have a positive average impact when students receive support and additional time before moving on.
How we use it
Impact varies with implementation, subject and student. Mastery is a design principle, not a guarantee.
Open the original sourceWhat it reports
The guidance emphasizes human accountability, age-appropriate design, privacy and maintaining meaningful human agency around educational AI.
How we use it
This is a governance framework, not an efficacy study.
Open the original sourceWhat it reports
The policy evaluates digital use through content, context, product design and what screen use displaces, rather than treating every minute as equivalent.
How we use it
Purposeful learning still needs boundaries, movement, relationships and responsible adult supervision.
Open the original sourceWhat it reports
Thailand's national mathematics, reading and science results provide context for the wider education environment.
How we use it
National samples do not describe a particular international school, family or child.
Open the original source