Randomness and time enter through explicit inputs rather than Math.random(), Date.now() or background timers, so a seed and command history can reproduce the same day.
Public · Economic simulation
Lemonade
A reconstruction of the classic Lemonade Stand built around three operating decisions, one day of sales and immediate economic consequences.
- Niche
- Accessible business simulation
- Core stack
- TypeScript · Lit · Three.js · Vite 8 · deterministic simulation
- Interaction
- Inventory · advertising · price · submit
- Presentation
- Vector 3D · procedural audio · accessible charts
System view
ConditionsThree decisionsDeterministic dayCertified ledger
The study
Lemonade keeps the daily operating surface to three decisions: inventory, advertising and price.
Weather and market sentiment act as conditions to interpret. The result is a compact business simulation that teaches demand, risk, diminishing returns and cash discipline through repeated play.
System structure
Three variables form the complete daily control surface.
Small control surface
Complexity appears through consequences, environment and progression.
Deterministic economic model
The simulation preserves the recognizable shape of the Apple II economics through typed, testable rules.
Progressive responsibility
Taxes, banking costs, interest and supplier economics extend the economic model while preserving the core daily decision loop.
Historical character
Vector 3D and procedural audio reference the charm of early computer games through an original contemporary presentation.
Product and interaction design
The interface presents the stand, weather and three daily decisions directly.
The neighborhood, stand activity, weather, customer flow and short audio motifs make the economic result feel like a day that happened. Charts and the ledger provide a second layer for reflection before the next decision cycle.
Keyboard play, precise numeric entry, reduced-motion modes, non-color cues and textual equivalents form part of the core interaction design.
Recent development
The simulation can demonstrate that its economy is repeatable, portable and reviewable.
Versioned persistence preserves phase and random-generator state, adds IndexedDB autosave, and supports import, export and reset while retaining deterministic replay.
A separate balance-certification layer runs fixed-seed strategies, checks accounting and gameplay invariants, and produces deterministic reports in CI. Game balance becomes reviewable evidence that can be compared across changes.
Reproducible persistence
A saved run restores both visible state and the random sequence that determines future days.
Certified balance
Reference strategies and demand probes catch regressions in difficulty, accounting and economic response.
Architecture lens
The economic simulation is deliberately unable to reach ambient platform state.
Simulation code cannot import UI, scene, audio, Tauri or Node layers and cannot call ambient random, clock or timer APIs. Those constraints keep the economic model deterministic, portable and independently testable while presentation remains free to evolve.
Lit UI, Three.js scenes, Web Audio, persistence and optional native packaging consume simulation output without becoming dependencies of the rules themselves.
Fixed-seed strategies and demand probes exercise accounting and balance invariants in CI, making economic regressions observable as evidence rather than subjective play impressions.
Mobile-first breakpoints, semantic dialog behavior, hover capability checks and layout anti-pattern rules are promotion gates rather than review preferences.
Development approach
The method begins playfully and gives useful discoveries rigorous support.
The architecture keeps deterministic accounting and demand rules in a pure package with injected randomness and fixed-precision money. Native DOM/SVG presentation, Three.js rendering, Web Audio, storage and optional Tauri capabilities remain adapters around that core.
A new weather curve, visual treatment, sound idea or unusual strategy can be tried quickly and replayed against the same seeds. The working loop is hypothesis → playable experiment → deterministic comparison → promote or revise.
AI-assisted development follows the same discipline: small reversible changes, explicit intent, fixed-seed tests and reviewable reports let rapid implementation amplify exploration while preserving the game’s economic character.