spr

To compress text for LLM-to-LLM transfer without losing semantic meaning, convert the source into a Sparse Priming Representation (SPR). This format uses dense markdown bullets to preserve structure and nuance, allowing a separate LLM to reconstruct the original material with high semantic recoverability.

A Sparse Priming Representation is a compact, informationally dense format designed for knowledge priming. It works by converting each paragraph into non-overlapping markdown bullet statements, enabling a downstream LLM to faithfully reconstruct the original scope and dependencies using minimal tokens.

You create a context handoff by generating an SPR from your source text. This produces a compact markdown output with one concept per bullet and a final 1–10 reconstructability rating, ensuring the downstream LLM can execute tasks with low drift.

Yes, this content compression method preserves the original markdown structuring. It applies semantic group headings and markdown subheadings to mirror the source's layout, ensuring structured organization is maintained alongside the compressed priming statements.

A key limitation is that sparse priming representations strictly enforce in-scope content fidelity, meaning they do not infer outside information. While they maximize semantic recoverability for multi-paragraph sources, the reconstruction accuracy depends on the downstream LLM's ability to interpret the dense bullets.