bmad-distillator

Lossless distillation compresses source documents into token-efficient distillates while preserving all facts, decisions, and constraints. This minimizes token usage for downstream LLM workflows without sacrificing information, utilizing a multi-stage workflow to analyze, compress, and verify outputs.

Round-trip validation verifies distilled LLM context by triggering end-to-end checks using the --validate option. This enforces frontmatter-driven metadata and required inputs to ensure the compressed distillate remains completely lossless compared to the original source documents.

Yes, flexible outputs allow generating either a single distillate beside the primary source or a folder of semantically partitioned, sectioned distillates. This organizes dense LLM context into manageable partitions while maintaining maximal token efficiency.

Yes, the distillation workflow enforces frontmatter-driven metadata and required inputs for LLM contexts. This ensures structured tracking of the lossless compression stages, from initial document analysis and compression to final verification and optional round-trip validation.

Lossless distillation for dense LLM context requires providing one or more source documents to function. While it maximizes token efficiency, the multi-stage analysis, compression, and verification process may introduce processing overhead compared to simple truncation or basic summarization methods.