Hi HN — I built a bit-for-bit deterministic analytics engine.
It runs classical ML pipelines (normalization → canonical transform → deterministic K-Means) with zero nondeterminism:
• no floating-point divergence
• no randomness
• no environment drift
• no timestamp or locale sensitivity
• Docker-pinned numeric behavior
• reproducible across machines, OSes, and hardware
The OSS drop includes:
• deterministic ingest + normalization
• deterministic K-Means (Iris + Wine)
• golden-reference hashes
• cross-machine reproducibility tests
• 3-machine ingest demo video
(direct download: https://github.com/bryanziehl/prima-veritas/releases/downloa...
)
• MIT license + full docs + architecture diagrams
If you work in ML, science, infra, or compliance, you already know how painful nondeterministic pipelines are.
This project is a first “Hello World” toward a broader deterministic verification kernel.
Feedback, critique, or reproducibility tests welcome — especially on different machine architectures.
Happy to answer anything live.
It runs classical ML pipelines (normalization → canonical transform → deterministic K-Means) with zero nondeterminism:
• no floating-point divergence • no randomness • no environment drift • no timestamp or locale sensitivity • Docker-pinned numeric behavior • reproducible across machines, OSes, and hardware
The OSS drop includes:
• deterministic ingest + normalization • deterministic K-Means (Iris + Wine) • golden-reference hashes • cross-machine reproducibility tests • 3-machine ingest demo video (direct download: https://github.com/bryanziehl/prima-veritas/releases/downloa... ) • MIT license + full docs + architecture diagrams
If you work in ML, science, infra, or compliance, you already know how painful nondeterministic pipelines are. This project is a first “Hello World” toward a broader deterministic verification kernel.
Feedback, critique, or reproducibility tests welcome — especially on different machine architectures. Happy to answer anything live.