The Sound project, available on GitHub, is an experimental programming language designed to ensure byte-level determinism in computations. Unlike traditional languages developed by corporations or academia, Sound focuses on solving the problem of non-determinism across different hardware architectures like Intel and ARM.

The core of the language, SCore v0.3, utilizes A-Normal Form (ANF) to fix the order of instructions within a graph, preventing compiler-driven reordering. To maintain consistency, Sound uses strict IEEE-754 64-bit float serialization, avoiding extended x87 registers. The system relies on the Golden Trace concept, where an execution log serves as a mandatory specification.

Sound is structured in layers: a declarative layer for system design, a functional layer for algorithms without side effects, and a foundation layer supporting six data types. Errors are treated as values rather than exceptions, allowing them to propagate through the graph. For interoperability, the language uses the TRIOMAT binary protocol.

The language is designed for practical applications in embedded systems, fintech, and scientific research. It generates C99 code that operates without malloc, suitable for environments like STM32. In finance, it ensures identical calculation results across different database versions, while in science, it guarantees reproducible research results regardless of the hardware used.