A sentence in the front matter is a claim. A prompt root inside the edition root is a commitment anyone can test.
Publishers are being asked whether a model was involved in a work and how. The usual answer is a line of front matter. LPS-1 gives a better one: when a model is used, the prompts are hashed into their own Merkle tree, and that prompt root is one of the four roots inside the edition root. The edition also records the model name and a hash of the prompt set.
The disclosure becomes a commitment. If someone later claims the prompts were different, or that no model was used, the root will not match. A human-only edition has an empty prompt tree and says so. A model-assisted edition has a prompt tree and says so. Either way, a reader can know, and can check.
This is the same discipline UnyKorn applies to its own reporting: every figure in the annual report family resolves to a ledger row with the command that produced it. A disclosure you cannot check is a sentence. A disclosure you can check is evidence.
Insight · 2026-09-09 · sourced to the evidence ledger and the chain