Deterministic reliability stack for LLM pipelines

I have been spending the last few months wiring up a deterministic reliability stack for structured LLM pipelines.

Today, LLM Contract Check (locc) and Release Governor went live on PyPI. EGA went live last week.

The stack is straightforward:
LLM Contract Check - CI contract testing to catch schema regressions.
Release Governor - Blocks staging promotion if malformed outputs leak.
EGA - Runtime enforcement. Forces outputs to ground against source evidence before they move downstream.

The idea is simple:
don’t wait until production logs or human evals tell you something broke.

Try to catch:

• unstable contracts in CI
• leakage before deploy
• unsupported outputs at runtime

Still early.
Not benchmarked.
Definitely not claiming this "solves AI safety."

I'm mainly looking for engineers building RAG or structured-output systems who are willing to plug pieces of this in and tell me where the assumptions break.

pip install llm-locc
pip install llm-release-governor
pip install ega