MODEL INDEPENDENCE ANY LLM

Built to serve any LLM.

The family is model-agnostic by construction: it hooks the harness's token output stream, not the model. Whatever LLM produced the text - DeepSeek, GLM, or the next one - the same six stream flavors and two governance roles apply, because unicode drift and unpinned manifests are model-independent problems.

The reference models

BATTLE-TESTED ROUTES
Engine selection: DeepSeek V4 Flash (120 tps) GLM 5.3 Flash Raw LLM Stream
DeepSeek V4 Flash VERIFIED STREAM

The harness route the family is developed and smoke-run against - every flavor verified on its token output streams.

GLM 5.3 Flash VERIFIED STREAM

The second battle-tested route - the same normalization and governance pipeline holding on a different vendor's stream.

What applies to every model

7 NORMALIZERS
DASH QUOTES ELLIPSIS SPACES INVISIBLE FULLWIDTH FILE
Model Independence

Normalizers sit on the stream, not on the model.

Extensively battle-tested on DeepSeek V4 Flash & GLM 5.3 Flash token output streams. The dash, quotes, ellipsis, spaces, invisible and fullwidth flavors run on every token batch before it reaches your files; the file flavor re-normalizes what is already on disk; the governor and pinner keep every manifest pinned - regardless of which engine wrote the output.

@playform/dsh-family: the complete deterministic stream governor matrix.

@playform/hooks: zero runtime overhead normalizers with zero unicode drift.

@playform/runtime: pinned across 12 packages under atomic version gates.