The Pioneer story, explored one modeled learner at a time. Move through 34,300 synthetic records across 2024–2026. Follow the path from a first idea to a live AI system.
Every figure on this page comes from a synthetic dataset. The 34,300 student records were generated (NumPy PCG64, seed 20270101) for program design, evaluation planning and dashboard prototyping. No record is a real child, site or app, and none of these numbers are observed program results.
Trace the cohorts, reveal the learning lift, or inspect every stage. Your filters carry through.
Six signals recalculated every time you change a filter. Each one links to the chart behind it.
Each of the 34,300 students in view is one strand threading nine measures. The paths shared by more modeled learners glow brightest. Drag along an axis to brush a range, drag a title to reorder, click a title to flip, or untangle the field automatically.
A deployment-first sequence makes the student's first artifact a live URL. Each bar is the share of students who reached at least that stage; colors follow the credential tier a stage earns.
Ribbon width is the number of students. Ribbons take the color of the credential tier they carry, so the brightest streams are students who reached Tier 4 AI Pioneer.
k-means++ clusters every student on ten standardized measures, and a principal component analysis projects them into three dimensions. No cluster is labeled by hand: each name comes from that cluster's two most distinctive measures. Select an archetype to filter the whole page to it.
Enrollment by cohort with the partner sites and jurisdictions active that year. Outcome rates rise as the curriculum, facilitators and managed hosting mature; that rise is a model assumption, not a finding.
Its records are held in a private vault and have a planned release date of Tuesday, March 30, 2027. Every figure on this page covers 2024–2026 only.
Each state rises by the number of students in view and glows by the metric you pick. Tap a state to filter the whole page to it. Site IDs are coded and do not map to real clubs.
Every partner site is a six-petal glyph. Sites are laid along a generalized Hilbert curve, so sites with similar outcome profiles sit next to each other. Click a site to filter the whole page to it. Site codes are synthetic and do not map to real clubs.
Every app counted here resolved at a public URL. Hosting shares show the modeled shift toward ZEN Arsenal managed hosting across cohorts.
Each ridge is the full score distribution for one construct, before and after the program. Within-subject pre/post with no comparison group, so gains are descriptive rather than causal.
Where students land on two measures at once, shown as a landscape. Height is the density of students in view, contours mark equal density, and the surface morphs whenever a filter changes. Change mode shows how the landscape shifted from the first cohort to the latest.
Outcome rates by student characteristic against the average for everyone in view. Home broadband is modeled as the largest structural constraint on a deployment-first program.
Tier follows the highest stage reached. Anchor status reflects on-chain confirmation of the issued badge.
Every row is a synthetic record. Search by ID, sort any column and open a record to see that student's modeled journey from enrollment to capstone.
ZEN AI Co. launched the AI Pioneer Program with the Boys & Girls Clubs of Greater Washington as the first youth AI literacy program in U.S. history. 2027 marks the fourth consecutive year of the partnership, with outreach expanding to YMCA chapters, Big Brothers Big Sisters and other youth-development partners.
Students ages 11–18 build AI-powered, cloud-hosted agents and publish them on Hugging Face, GitHub and ZEN Arsenal, ZEN's own agentic builder platform. Prompting, data, auth, ethics and telemetry are taught against software real users can already reach.
Progress is recorded as Open Badges 3.0 credentials anchored on NEAR Protocol, from Tier 1 Deployer to Tier 4 AI Pioneer. A deployment-first program leaves machine-checkable evidence: a URL that resolves, an uptime number, a user count.
Drawn from the dataset's own documentation. These figures must not be presented as observed program results in any grant application, federal proposal, press release or investor material.