Choose your household
All households start with 20 credits. Geographic labels denote hypothetical scenarios, not measured differences or inherent traits. Changing a selection resets the run. Future variable income is not displayed during play; the synthetic schedule is inspectable in the rules.
Round 1 of 6
Goal: earn satisfaction points, Σ√(units consumed). Saving may smooth consumption. No borrowing or interest; leftover money after round six earns no points. This score is an illustrative preference assumption, not measured well-being.
Your run
| Round | Income | Spent | Units | Saved |
|---|
Controlled comparison
Run all three access scenarios with the same income timing and spending rule. The benchmark spends 80% of available credits in rounds 1–5 and everything in round 6. It is a heuristic, not an optimal strategy or observed human behavior.
No comparison run yet.
| Scenario | Total units | Score | SD | Final savings |
|---|
Access improvement is an exogenous reduction in delivery/access costs, not a funded subsidy. Its implementation cost is unknown; these results cannot rank policy cost-effectiveness. SD is the population standard deviation across six rounds; lower is smoother, not necessarily better overall.
Inspectable rules, assumptions & sources
Author: Zhenning Wang (zw317). Model version 1.0, prepared September 6, 2026. Stable incomes: [100,100,100,100,100,100]. Variable incomes: [40,160,40,160,40,160]. This fixed sequence is a reproducible cash-flow stress test, not a stochastic model of income uncertainty.
For round t: available = previous savings + income[t]; spending = units × price; next savings = available − spending. Units must be finite, nonnegative, and affordable. Reset clears all choices and restores 20 initial credits. Prices and incomes are hypothetical. Manual decisions and benchmark runs use the same transition function.
The current prototype implements budget dynamics, not trained AI, causal identification from real data, an equilibrium model, or a full 2056 system. Future work can collect consented decisions and audit prediction errors by location. No player data are sent, saved persistently, or collected by this game.
Conceptual foundations: Kappes et al. (2023), consumption smoothing; Couture et al. (2021), rural e-commerce access; Aiken et al. (2023), urban–rural prediction bias. These sources do not calibrate the toy parameters.
Static HTML/CSS/JavaScript only. No API calls, external libraries, database, backend, keys, secrets, or paid service. Hosting, network delivery and browser execution still consume resources.