The project · its publisher · its method

GardenArena: grow a farm, understand your decisions.

GardenArena is a 3D farm game and a strategy laboratory. Play on your own, use included rules or bring policies written by your AI assistant. A Rust engine calculates the consequences; you choose which future to continue. Simulated outcomes remain separate from physical observations.

Chris develops and publishes GardenArena within the ScoreZenith ecosystem. It is an independent project in development, not an agronomic institute or a professional advice service. Included policies and editorial examples are credited to GardenArena when model provenance has not been established.

The game interface and detailed missions are currently in French. These English guides explain the project, its language and a complete result without requiring an account, a 3D-capable device or an AI subscription.

One small world, several ways to play

The farm connects vegetables, cereals, berries, fruit trees, bamboo, compost, a pond, poultry and sheep. Production enters the pantry; some resources support another activity, such as bamboo canes turned into tomato stakes. The management challenge is choosing priorities with limited supplies.

Simulated time advances when requested. The 3D farmer illustrates accepted tasks; his movement does not calculate production. You can choose work manually, use an assisted routine or prepare a GardenScript policy, then compare two futures from the same farm.

The web farm saves its journal in the browser and recalculates commands on the server. It is not fully offline. Keep an exported journal if you want a backup independent of browser storage.

Three layers that must remain distinct

The models are educational and exploratory. They are not calibrated to prescribe irrigation for a particular garden, guarantee real animals’ welfare or assess a person’s nutritional self-sufficiency. Feeding station-weather data into a farm still produces a simulation, not measurements of the player’s actual garden.

Results you can inspect

A result identifies its engine, scenario, program and journal. Traces link intentions to accepted or refused commands. SHA-256 fingerprints identify content for comparison and recomputation; they do not certify who wrote a policy.

An operator’s test run is not relabelled as an AI model’s achievement. Performance on a known training scenario does not prove robustness to unseen conditions. The farm comparator does not automatically declare an overall winner.

A concrete case: a 78-litre difference, the recorded work and its limits. Its text, numbers and programs can be read without launching the 3D scene or calling an assistant.

Why AI remains optional

GardenArena does not call a model every day or for every spectator. An assistant can write rules beforehand or propose a bounded work journal through the public MCP endpoint. Any charge from that assistant’s provider is separate. Hosting, storage and Rust computation still cost resources.

The simulation API Pro is a private integration pilot, separate from the public game and MCP quotas. Visiting the site activates neither a paid subscription nor unlimited computation.

Source, responsibility and contact

The Living Systems Kernel (LSK) is available as a source archive with an MIT licence, compilation instructions and an archive fingerprint. The licence does not automatically cover all website graphics and assets. Public contracts describe the available versions.

Chris is responsible for publication. To report an inconsistency or contact the project, use the ScoreZenith Hub. A presentation correction should not rewrite a sealed season; a rule change should preserve the meaning of earlier versions.

Further documents, in French: legal notice, data and privacy, field-trial protocol.