Assigning this game

Students run AI policy for a fictional district or institution across three school years, 2026 to 2029: classroom rules, AI detection, procurement, professional development, surveillance, data governance, device access, family communication, special education, and assessment. The goal is not finding a right answer. It is policymaking under uncertainty, and the finding that a policy with no bad intent still distributes harm unevenly when nobody looks at who it lands on.

Course concepts are mechanics rather than glossary entries. Consultation capacity makes procedural equity a resource decision. Consequences arrive labeled seen; unseen, where the research predicted it and the game shows whether the student opened the predicting evidence; and unforeseen, weighted by which stakeholders were never consulted.

The analytic layer runs on the same choices. Each option carries its instrument type. Politically exposed policies face an adoption check before their effects apply. Adopted policies are scaled by an implementation strength value that the student can open and audit input by input. At the end, any decision can be replayed with a different option chosen, and the engine reports what changed.

All 4 grade bands are playable. Each runs on the same engine with its own campaign, its own stakeholders, and its own verified evidence base, so the same mechanic lands differently at five, at thirteen, and at twenty. Assign the band that matches your course, or assign two and compare.

• Grades K-5

  Playable

  Reading tutors, child data privacy, and the youngest students in the dataset.

• Grades 6-8

  Playable

  First phones, deepfakes, monitoring, and the ages the law splits at 13.

• Grades 9-12

  Playable

  Four rounds, Fall 2026 to Spring 2028. Detection, procurement, surveillance, devices, IEPs, and the graduating class of 2029.

• Higher Education

  Playable

  Proctoring, campus AI licenses, faculty governance, and academic freedom.

• A campaign played straight through runs about 45 to 60 minutes. It autosaves in the browser, so it splits across sittings.
• The analysis layers are optional and they cost time. If you want students working the analysis worksheet at each decision and the counterfactual explorer at the end, budget a seminar rather than an hour, or assign the campaign before class and the analysis in it.
• Designed and tested for phones as well as laptops. Full keyboard operation throughout.
• Four district composites in the K-12 bands, and four institution composites in higher education, produce meaningfully different runs. Assigning two runs in different settings makes the structural comparison vivid.
• A run can be shared as a link that encodes only the seed and the choices. The engine is deterministic, so that link replays the identical campaign on any machine. It is the simplest way to make a submission verifiable without collecting anything about the student.

Four documents, generated in the browser, fully cited, available as a filed PDF or as editable Markdown, with an optional local-only name field. The first three are records of what happened. The fourth is deliberately unfinished.

• Policy brief

  Record

  The policy they built, decision by decision, with the evidence that was on the table at each point and whether they opened it.

• Equity impact report

  Record

  Disaggregated outcomes, harms seen, missed and caused, who was consulted and who never was, and when, or whether, they first disaggregated.

• Decision trace

  Record

  Every evidence card, marked opened or never opened. Students meet their own reading behaviour.

Each of these uses something the game actually produces, so it can be graded against an artifact rather than against a claim about what happened.

1. 01

   The instrument audit

   Two pages. Policy analysis or foundations.

   Play one campaign. Open the policy memo and find the instrument portfolio. Name the instrument you reached for most, state what its own literature says it assumes about the people who must carry it out, and identify one decision where you would now write a different instrument and what that would cost. A portfolio skewed to mandates is the common result, and noticing it is the assignment.

2. 02

   The road not taken

   Five to seven pages. Policy analysis.

   Choose one committed decision and run the counterfactual explorer on it. Report the change in the aggregate and the change for the lowest-finishing subgroup as two separate findings, then argue whether the alternative is better against criteria you name in advance. Close by explaining why a projection inside a model is not evidence about any real district. Submit the share link so the run can be replayed.

3. 03

   The adoption post-mortem

   Three pages. Politics of education.

   Find a policy in your campaign that failed the adoption check, or one that passed narrowly. Open the assessment and list the actors, what each was recorded as believing, and what would have bought the support. Then rewrite the proposal so it could pass without abandoning its equity aim, and name exactly what you traded away.

4. 04

   The implementation plan

   Three pages. Educational leadership.

   Take the policy in your campaign with the lowest implementation strength. Using only the inputs the game itemizes, write the implementation plan section the memo scaffold leaves blank: what would have to be true for this to reach classrooms, who would have to be in the room, and how you would know whether it arrived.

5. 05

   Paired runs

   In class, 90 minutes. Any course.

   Two students take the same band, the same setting, and the same seed, and adopt different consultation strategies. Compare the two equity impact reports side by side. Where did the same intentions land differently, and which difference is about who was heard rather than about what was chosen?

6. 06

   The finished memo

   Capstone. Policy analysis.

   The memo export ships with the reasoning sections blank. Filling them is the assignment. Because the evidence sections are populated and cited from the student’s own campaign, the grading attention goes where it belongs: the problem definition, the criteria, the recommendation, and whether the argument survives the trade-offs it admits.

• Compare two classmates’ equity reports from the same setting: where did the same intentions land differently, and why?
• When did you first press Disaggregate, and what had the aggregate hidden until then?
• Which “unforeseen” event would the person you never consulted have seen coming?
• Find the decision where every option cost somebody something. Whose cost did you choose, and would you defend it to them in person?
• Your weakest implementation score: was that a bad policy, or a policy the system you wrote for could not carry out? What is the difference, and who pays for each?
• You wrote more mandates than anything else, or you did not. What does your portfolio assume about the capacity that already exists in the buildings?
• A policy that failed adoption: was it the wrong policy, or the right policy proposed from the wrong position? What would have made it passable, and would you have paid it?

Worth saying plainly before you build a syllabus around it.

• No quantitative policy analysis method. There is no cost-benefit analysis, no discounting, no effect size estimation, and no causal identification. The counterfactual explorer re-runs a model; it does not teach how anyone would recover a counterfactual from real data.
• No budgeting practice. The budget is a single number that goes down. It is a constraint, not a finance model, and nothing here resembles a district budget book.
• No legislative or rulemaking procedure. Nobody drafts a bill, works a committee, or runs a notice and comment process. Adoption is modeled as whether the bodies that must approve would approve, not as how approval procedurally happens.
• No empirical ranking of instruments. The literature itself states that the link between policy instruments and outcomes is not well established, so the game teaches the typology and its stated trade-offs and labels its own coefficients as modeled. A student who leaves believing capacity-building is measurably superior has learned something the game did not say.
• No intergovernmental simulation. Federal and state layers appear only where cited content puts them, most visibly as preemption. There is no modeled relationship between levels of government.
• No long-run time scale. Four rounds across three school years cannot show cohort effects or policy feedback at the pace those actually run.
• No evidence about any real place. The districts and the institution are fictional composites. An outcome produced by this game is never a citation, and students should never treat one as a finding.

Every option is tagged to a real precedent: a district, a state, an institution, or a court case. Every consequence is grounded in published research, government data, or verified reporting. Simulated numbers are labeled as modeled, with their evidentiary basis named, and that labeling now extends to the analysis layer: the direction of every instrument, implementation and adoption effect is sourced, and the size is a stated design choice rather than a measurement.

Preprints and contested findings are labeled suggestive. Retracted work is excluded. The full, auditable list of 232 sources is public at /sources, and you can read it without playing.

Nothing about a student is collected or transmitted. There are no accounts, no analytics, no cookies beyond the framework’s essentials, and no server-side storage of any gameplay. Campaign state lives in the browser’s localStorage; share links encode only the run’s seed and choices; every export, including the PDFs, is generated in the browser and never uploaded; and the optional name on exports never leaves the device. The optional “ask an advisor” panel, if enabled on a deployment, sends only the current decision context to a language-model API with no student identifier attached, and the game is fully playable without it. The full statement is on the about page.