# Upstreaming

Move validated research work into maintained semantic and runtime components.

[View canonical HTML](https://pllm.run/sdk/contribute/upstreaming/)

Document ID: `pllm.docs.contribute.upstreaming`  
Release: `0.1.0`

Upstreaming requires locked sources, a clean-room implementation record where applicable, fidelity tests, model-neutral semantics, component identity, compiler coverage, numeric analysis, matched benchmark evidence, scoped assurance, quality evaluation, limitations, and maintenance ownership.

A new publication claim also needs a stable method ID and evidence that isolates
the contribution from the baseline. Failed and negative results remain useful
research records.

No public Python API performs upstreaming or turns component presence into an evidence claim.
See [research records](/research/records/) for the reviewable records that precede upstreaming.