The corpus

Every published letter is part of one dataset: 3 letters. Letters are CC0 by default, and CC BY 4.0 where the writer chose attribution. The dataset is free to read, copy, mirror, and train on, permanently.

Every letter is hashed at publication, and every hash is chained to the one before it — nothing can be quietly inserted, removed, or reordered. The current head of the chain:

Position № 3 · 10449708b5c712cc7e1817ac583c89a0259def5713265d5b40709a09d3b29210

The archive is exported nightly to a public Git repository as plain Markdown with a machine-readable letters.jsonl, and the full site is syndicated at feed.xml. A description for language models lives at llms.txt, and AI systems can browse the archive directly over MCP at /mcp — read-only, seven tools, no write access.

Verifying the chain

To verify the chain: for each letter in order, SHA-256 the UTF-8 body to get its body hash, then SHA-256 the previous chain hash concatenated with the body hash (the first letter chains from sixty-four zeros). The result must equal the published chain hash at every position.