Show HN: Jacquard, a programming language for code written by AI and reviewed by humans

Jacquard is a research project from FriendMachine designed to execute, review, simulate, and trust programs written by models and reviewed by humans. It is currently provided as a research prototype at the 0.1 release stage.

Unlike existing languages, this language offers the following core features:

• Effects exposure: The language explicitly declares at the language level what external effects a function may cause, such as network communication, and unauthorized effects are rejected at runtime.
• Multi-environment simulation: The same code can be run across various environments (Handlers) such as a real network, a virtual script, or a probabilistic model, to test agent behavior.
• Probabilistic reasoning: For finite discrete models, exact probabilities can be computed, allowing possible patches and their probabilities to be derived when a test fails.
• Canonical Identity: Code with the same structure is recognized as identical code even if comments or formatting differ, preventing unnecessary test re-runs.

The project aims to let human reviewers instantly grasp what code written by an AI agent can touch, and how certain it is, without having to read through the entire code.