Prepare WordStar Archives for AI and RAG — Offline

Why not PDF for RAG

PDF extraction reintroduces layout noise. Markdown from a native WordStar parse keeps headings and emphasis as structure a chunker can split on, without wrap-flag garbage.

The offline prep path

• Batch-convert `.ws` / `.wsd` / `.ws5`–`.ws7` to Markdown on Windows.
• Keep originals offline; feed only the Markdown tree to your indexer.
• Chunk on headings and letter boundaries, not on raw 2k-character windows through mojibake.
• Embed with your local or VPC model stack — never paste privileged letters into a consumer chat box.

What the converter must fix first

Retrieval-Augmented Generation (RAG) quality collapses if the source text still contains high-bit wrap bytes and literal `.pl 66` lines. Unwrap and interpret those layers once, offline, then treat Markdown as the system of record for embedding.

Privacy boundary

Conversion is local. Your embedding service is a separate decision. Do not blur them: get clean Markdown on this PC first, then apply whatever AI policy your organization already uses for other documents.