CASE STUDY · PUBLIC
scorestrata
A deterministic music compiler: a plain-language brief in, a mastered WAV out, byte-identical on replay from the same contract and seed.
Status ruling active development
Problem
Media generation tools treat output as a roll of the dice: the same prompt produces a different artifact each run. Composition for production needs the opposite property.
Build
- A contract-first compiler: the brief is a typed contract, and the seed is an explicit input, so the render pipeline is a function rather than a sampler.
- The mastering chain implemented in Rust, so byte-identity is a property of the code, not of a pinned external tool.
Outcome
- 944 tests passing with zero failures at commit c5dc71b, re-measured on a clean export. The README badge said the same number: the one badge the source audit found exactly right.
- Determinism verified by replay: the same contract and seed produce a byte-identical WAV, which is what makes the tool usable in a pipeline rather than a toy.
Provenance
| Gate, executed | 944 passed, 0 failed | cargo test at c5dc71b, 2026-10-20 |
|---|
Reproduce
Clone the repository, check out c5dc71b, and run `cargo test`: the runner output re-derives the figure. Render twice from one contract and compare bytes for the determinism claim.
https://github.com/ishan-parihar/scorestrata
Published documents
From the documentation
A typed brief compiles to byte-identical WAV output with the seed as an explicit input, which makes every render a reproducible artefact rather than a generation.
architecture.md at cf8ce6f (2026-10-21)
The architecture document is a compiler contract, not a description of a model: brief in, deterministic PCM out, with the composition rules stated as the pipeline.
architecture.md at cf8ce6f (2026-10-21)
Determinism is the selling point and the document states it as the contract, which is why the same brief and seed reproduce on any machine.
architecture.md at cf8ce6f (2026-10-21)