ZENLLC/ai-pioneer-program-modeled-outcomes

AI Pioneer Program: Modeled Outcomes, 2024–2026

MODELED DATA. Every record in this dataset is synthetic. The 34,300 student records were generated, not observed. No row corresponds to a real child, site, facilitator or app. These figures must not be presented as observed program results in any grant application, federal proposal, press release or investor material. Label every chart derived from this data "modeled".

The 2027 cohort is sealed until Tuesday, March 30, 2027. Its records, the 2026 cohort's retention into 2027, and the four-cohort source workbook are encrypted in this repo. The key is published here on that date and the explorer unlocks the cohort automatically.

Interactive 3D explorer: ZENLLC/ai-pioneer-outcomes-explorer · Program: zenai.world/ailiteracyyouth · ZEN Arsenal: arsenal.world

Summary

This dataset models what ZEN AI Co.'s AI Pioneer Program could look like at scale. ZEN AI Co. runs the program with the Boys & Girls Clubs of Greater Washington and describes it as the first youth AI literacy program in U.S. history; 2027 is its fourth consecutive year. Students ages 11–18 build AI-powered, cloud-hosted agents and publish them on Hugging Face, GitHub and ZEN Arsenal, ZEN's agentic builder platform.

The dataset exists for program design, evaluation-plan drafting, grant narrative structure, dashboard prototyping, and IRB and evaluation-vendor scoping.

Files

File	What it is
roster.parquet	The 34,300-row student roster for the 2024, 2025 and 2026 cohorts, 51 columns. This is the default config the dataset viewer and datasets load.
AI_Pioneer_Student_Roster_2024-2026_SYNTHETIC.csv	The same roster as CSV.
viewer_app.html	The interactive 3D explorer that the Space loads.
viewer_meta.json, viewer_rows.pack	Compact columnar build of the 2024–2026 roster that the explorer reads.
viewer_sealed_2027.pack, sealed_2027_originals.zip.enc	The 2027 cohort and the four-cohort source files (roster, CSV, workbook), encrypted with AES-256-GCM. They cannot be read until the key is published on March 30, 2027, when the explorer unlocks the 2027 cohort automatically.

Load it

from datasets import load_dataset

ds = load_dataset("ZENLLC/ai-pioneer-program-modeled-outcomes", split="train")
df = ds.to_pandas()

import pandas as pd

df = pd.read_parquet("hf://datasets/ZENLLC/ai-pioneer-program-modeled-outcomes/roster.parquet")
deployment_rate = (df.highest_stage >= 2).mean()  # 0.824

Text values such as None, n/a and n/a - no deployment are category labels, not missing values. If you read the CSV with pandas, pass keep_default_na=False, na_values=[""] to keep them.

The curriculum being modeled

The program is deployment-first. A student's first artifact is a live, publicly reachable URL, not a worksheet or a local file. Prompting, data, auth, ethics and telemetry are then taught against software real users can already reach. That produces machine-verifiable evidence: a URL that resolves, an uptime number, a deploy count, a user count, and an app that is or is not still live 90 days later.

Stage ladder

Stage	Definition	Credential tier
S0	Enrolled, no environment	None
S1	Cloud environment provisioned	None
S2	First live deploy (public URL)	Tier 1 · Deployer
S3	AI feature integrated in production	Tier 1 · Deployer
S4	Persistent data + auth layer	Tier 2 · Builder
S5	Public launch + custom domain	Tier 2 · Builder
S6	Iterated on real user telemetry	Tier 3 · Operator
S7	Capstone reviewed + credential issued	Tier 4 · AI Pioneer

Credentials are modeled as Open Badges 3.0 anchored on NEAR Protocol. Build tracks are Agent Builder, Data & Dashboards, Creative Studio and Civic Tech.

Headline figures (modeled)

Cohort	Students	Shipped a live URL (S2+)	Public launch (S5+)	Capstone (S7)	Withdrew (S0)	Median days to first deploy	Mean gain	Sites	Jurisdictions
2024	6,200	74.2%	47.8%	38.6%	16.8%	8	+21.6	58	16
2025	11,300	81.5%	56.6%	47.1%	14.1%	5	+23.7	104	30
2026	16,800	86.1%	63.8%	53.8%	11.8%	3	+25.5	163	40
2027	Sealed until March 30, 2027
2024–2026	34,300	82.4%	58.5%	48.9%	13.5%	4	+24.2	163	40

Across 2024–2026: 28,276 students shipped a live URL, 63.9% of shipped apps were still live at 90 days, apps reached 2.0M unique users in 30 days, and 26,582 badges were anchored on NEAR. The pooled-SD standardized assessment gain is 1.52, with no comparison group.

Schema

Column	Type	Description
student_id	string	Synthetic ID, AIP-{cohort}-{sequence}
cohort_year	int	2024, 2025 or 2026 (2027 sealed)
enrollment_date	string (ISO date)	Between Sept 9 and Nov 10 of the cohort year
age_at_enrollment	int	11–18
age_band	string	11-12, 13-14, 15-16, 17-18
gender	string	Female, Male, Non-binary / self-describe, Not reported
race_ethnicity	string	Eight categories including Not reported
frl_eligible	string	Free or reduced-price lunch eligible: Yes / No
locale	string	Urban, Suburban, Rural
home_broadband	string	Yes / No
state	string	Two-letter code; 40 jurisdictions including DC
region	string	Census region
site_id	string	Coded partner site (network-state-sequence). Not a real location
partner_network	string	Boys & Girls Clubs, YMCA, Big Brothers Big Sisters, School District Partner, Public Library System
delivery_mode	string	In-person, Hybrid, Virtual
facilitator_id	string	Coded facilitator
track	string	Agent Builder, Data & Dashboards, Creative Studio, Civic Tech
sessions_offered	int	24 (12 weeks, two sessions a week)
sessions_attended	int	0–24
attendance_rate	float	Attended ÷ offered
status	string	Completed, Partial - deployed, Partial - no deploy, Withdrew
highest_stage	int	0–7, see the stage ladder
stage_name	string	Label of the highest stage
deployed_live_url	string	Yes when highest_stage ≥ 2
public_launch	string	Yes when highest_stage ≥ 5
days_to_first_deploy	float	Days from enrollment to first public URL; empty when not deployed
total_deploys	float	Production deploys; empty when not deployed
app_category	string	One of 17 categories, or n/a - no deployment
hosting_stack	string	ZEN Arsenal (managed), Vercel + Supabase, Netlify + Serverless, Cloudflare Pages + Workers, Replit Deployments, or n/a - no deployment
app_uptime_30d_pct	float	30-day uptime percentage
app_users_30d	float	Unique users in 30 days
app_live_at_90d	string	Yes, No, or n/a
portfolio_published	string	Yes / No
pre_Found, post_Found	float	AI Foundations, 0–100
pre_Prompt, post_Prompt	float	Prompt & Context Engineering, 0–100
pre_DeployOps, post_DeployOps	float	Build & Deploy Ops, 0–100
pre_DataPriv, post_DataPriv	float	Data & Privacy, 0–100
pre_Ethics, post_Ethics	float	AI Ethics & Governance, 0–100
pre_total, post_total	float	Unweighted mean of the five constructs
score_gain	float	post_total − pre_total
credential_tier	string	None, Tier 1 - Deployer, Tier 2 - Builder, Tier 3 - Operator, Tier 4 - AI Pioneer
badge_issued	string	Open Badges 3.0 badge issued: Yes / No
badge_anchor_status	string	Anchored, Pending anchor, or n/a
mentor_hours	float	Facilitator and mentor contact hours
retained_next_year	string	Yes, No, or n/a - current cohort. While 2027 is sealed, the 2026 cohort shows n/a - current cohort

How the data was generated

• Generator. NumPy PCG64, seed 20270101. Re-running the generator reproduces the file exactly.
• Latent readiness. Each record carries a readiness score built from attendance, delivery mode, home broadband and age band plus Gaussian noise, mapped onto the stage ladder so realized cohort rates land on the model's targets.
• Program maturity curve. Outcomes improve cohort over cohort (74.2% → 86.1% deployment from 2024 to 2026) to reflect curriculum iteration, facilitator experience and a shift toward managed hosting. This is an assumption, not a finding.
• Equity structure. Rural students and students without home broadband are modeled with lower attendance and slower time to first deploy. The broadband gap is deliberately visible: it is the largest structural constraint on a deployment-first model.
• Assessment. Five constructs scored 0–100; the total is the unweighted mean. Build & Deploy Ops has the lowest baseline and the largest modeled gain.
• App telemetry. Uptime is modeled high because managed hosting is reliable. Weak projects show up as abandonment (not live at 90 days) rather than downtime. User counts are lognormal with a heavy tail.
• Hosting categories. This version models five hosting stacks. It does not yet break out Hugging Face Spaces or GitHub Pages as separate stacks.

Known limitations

• No counterfactual. Pre/post gains are within-subject with no control or comparison group, so they are descriptive rather than causal. A real evaluation needs a matched comparison group or a stepped-wedge design.
• Selection into enrollment. Students who opt into an AI build program are not a random sample of club members. Withdrawal is modeled; selection at the front door is not.
• Self-reported fields. Gender, race/ethnicity and broadband would be self- or guardian-reported in the field, with non-response. The Not reported level is understated.
• Site coding. Site IDs are coded and must not be mapped to real clubs or locations.
• Minor-subject data. A real version of this dataset is student data on minors. It would require guardian consent, a data governance policy, FERPA and COPPA review, and de-identification before any external sharing. Treat this schema as a governance planning artifact.

Intended uses

• Designing a program evaluation and its data collection instruments
• Drafting the structure of grant narratives and logic models, with every figure labeled modeled
• Prototyping dashboards and data pipelines before real data exists
• Teaching data literacy with realistic, privacy-safe tabular data

Out-of-scope uses

• Presenting any figure here as an observed program result
• Estimating the causal effect of the program
• Mapping site codes or records to real students, clubs or places

License

Copyright © 2026 ZEN AI Co. All rights reserved. The dataset is publicly viewable on Hugging Face for reference; no license to copy, modify, redistribute or reuse it is granted without prior written permission from ZEN AI Co. See LICENSE.

Citation

@misc{zenaico_2026_ai_pioneer_modeled_outcomes,
  title        = {AI Pioneer Program: Modeled Outcomes, 2024--2026 (Synthetic)},
  author       = {{ZEN AI Co.}},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/datasets/ZENLLC/ai-pioneer-program-modeled-outcomes}},
  note         = {Synthetic dataset. Not observed program results. The 2027 cohort is sealed until 2027-03-30.}
}