The founding paper

Soons, L. (2026). Agentic Stemmatics: Collation, Provenance, and Conservation for an Emerging Machine Textual Tradition. Zenodo (preprint, CC BY 4.0, not peer-reviewed).

• Concept DOI (always resolves to the latest deposited version): 10.5281/zenodo.22030516
• Deposited versions: v1.15 (20 Aug 2026) · v1.25 (24 Aug 2026)
• A plain-language introduction is at research.counterproof.io.

The paper states the mapping with its conditions attached: which classical operations transfer, which fail and why, and what replaces them. Its central corrections — the witness is a lineage sampled repeatedly, not a single output; shared errors carry no ancestry signal on present evidence, shared idiosyncrasies under a per-item improbability condition do — came out of adversarial review, and the paper keeps its revision history as data rather than housekeeping.

The paper practises the field on itself

Its drafting record is a worked example of the discipline it proposes:

• Adversarially reviewed across model lineages — more than a dozen review rounds by seats from four independent vendors, with findings, refusals and reversals recorded, and later revisions batch-gated by five seats before filing.
• A public, hash-chained closure log — github.com/clvstra/agentic-stemmatics-closure-log records review rounds, verdicts, and deposits as an append-only chain with a standalone verifier. Entry 1 is anchored to the Bitcoin blockchain.
• Version discipline — released versions are hash-pinned; a pinned version is never edited in place; corrections re-version. One violation of this rule occurred during drafting, was caught by the pin, and is recorded on the paper’s own face rather than erased.

None of that evidences the paper’s claims — surviving one’s own review process proves little, and the paper says so. It evidences the methods: this is what collation, provenance and conservation look like when applied to a living document.