| title | 02 · Product Boundaries |
|---|
- Configuration-driven game model
- Long-horizon time advancement (discrete events primarily, see the Time model)
- Player / actor simulation
- Combat (starting with analytic resolution)
- Resource economy
- Progression systems
- Player cohort segmentation
- KPI / metrics (formulaically defined, see KPI and metrics)
- Monte Carlo (multi-seed reruns)
- Parameter sweep
- Experiment result comparison
- Parameter sensitivity analysis
- Reproducible experiments
- JSON / CSV reports (Parquet behind a feature flag, see Data output)
Sandtable plans three run forms sharing the same Rust Core — there is no JS re-implementation:
| Form | Position | Phase |
|---|---|---|
| CLI | CI, large-scale experiments, developer automation | MVP (first delivery form) |
| Web (WASM + React) | Small-to-medium simulations in the browser, "try it in your browser" demos and teaching | Phase 6 |
| Desktop (Tauri 2) | Primary form for Windows and similar platforms: local files + native analysis, no Node / Python backend | Phase 7 |
Three shared constraints:
- Local-first: configuration, experiments, simulation, results, and analysis all happen locally — no account, server, database, or network needed;
- Single kernel: CLI / Web / Desktop all link
sandtable-core; writing a separate JS simulation for the web is strictly forbidden — otherwise "web results differ from desktop results" will happen sooner or later, and the reproducibility promise is broken; - Architecture first, features later: from Phase 1 the Core has been hard-constrained to be "compilable for WASM, reusable by Tauri, and decoupled from the analysis engine" (platform boundary in the Simulation kernel); Web / Desktop have landed per phases 6 / 7.
- Game engine Adapter (decided via delegation and landed, patrol R1, 2026-10-14): an engine-agnostic JSON contract in
crates/sandtable-engine(export + reference importer + backflow convention); decision record and spec in Engine Adapter design) - CI balance regression (decided via delegation and landed, patrol delegation, 2026-10-15): case KPI level assertions in the CLI tests; decision record in Direction draft §4)
- Visualization system graph & designer workflow (decided via delegation and landed, patrol delegation, 2026-10-15): program-generated system graph in chapter 04, designer workflow tour in chapter 21; decision record in Direction draft §4)
- Bayesian / GP optimization (on demand; Evolutionary / Pareto and the random-forest surrogate have landed, see Parameter sweep)
Keeping the project from losing control at the start:
- General-purpose MMORPG engine
- Game server runtime
- Full Machinations clone
- Unity / Unreal / Godot editor
- LLM players, reinforcement-learning players
- Complex behavior-tree editor
- Online SaaS, multi-user collaboration platform
- Complex automated-optimization UI
::: tip Core principle Get the Simulation Kernel and Experiment Loop right first, then build the product interfaces. :::
The PvP, guilds, and markets in the "not doing" list correspond one-to-one with the modeling assumptions in Project positioning: independent players, PvE only, economy as aggregates. These three assumptions together form the MVP's feasible set; items beyond them (inter-player interaction) are architecture-level changes, filed as follow-up directions rather than MVP patches.
The project ships under Apache-2.0, already established with the repo. Third-party dependency selection must consider license compatibility — for example evalexpr moved to AGPL-3.0 from 13.x, which is incompatible with Apache-2.0 projects, so it was excluded from formula-engine candidates (details in Configuration and formula engine).