AXN as Distributed Identifier — Vision, Verified State, and Work Plan v2.0 is a governance specification by Lee Sharks, dated 6 August 2026, establishing the architecture by which an AXN becomes a genuinely distributed identifier rather than a content-derived one held in a single archive.
Its governing claim is that no digital system makes every copy literally unerasable, and that the attainable goal is stronger than the impossible one: an identifier that remains meaningful, independently verifiable, historically complete and recoverable from any surviving node, with no domain, registrar, repository, account, platform, resolver, operator or storage network permitted to become its point of failure.
The document was synthesised from four blind drafts produced independently by ChatGPT, Gemini, Kimi and DeepSeek, then revised under five Assembly reviews. Its most consequential methodological choice is epistemic: model convergence is recorded as provenance, not evidence. Four substrates agreeing motivated the synthesis; the measured infrastructure and the dated failure record supply the evidentiary basis. To keep that distinction legible the document separates three kinds of statement throughout — VERIFIED, NORMATIVE and WORK — on the principle that a plan mixing measurement with intention cannot be audited.
Six normative invariants govern: the kernel is the identity boundary; every other layer is additive, with hex position and glyph seal serving as recognition rather than security; resolvers locate while hashes verify; history is never destructively rewritten; distribution requires independent custody under the LOCKSS test, so copies sharing one administrator, registrar, host or billing relationship constitute a single custody domain; and failure is a result rather than an absence. An explicit authority model distinguishes content authority, claim provenance, registry standing, current designation, custody evidence and historical fact — because a signature proves that a key made a statement and nothing further.
Five rulings resolve conflicts between the drafts. The most doctrinally significant refuses certification tiers and preservation grades, on the ground that AXN-SYMBOLON-SPEC §5.1 already fences the prestige format: publish custody facts, never grades, because facts inform where grades rank. The others cut rights assertions the Constitution does not make, forbid any mandatory transparency service, defer scheme registration behind working HTTPS resolution, and rule out optimising for a specific machine composer — since an observatory that games its own subject forfeits its standing.
Three capabilities are named as distinctive. Publication is a property distinct from validity, and must be instrumented at every node, because a peer serving 404s looks like redundancy while providing none. The lacuna is a headline capability rather than an error path: where a DOI returns an undifferentiated 404, an AXN returns identity, last known location and last verified hash — the hash being what makes a lacuna a recovery target rather than a bookmark. And measured persistence is held to be the differentiator no competitor can copy, since every other identifier system asserts persistence while this one is operated by an archive that measures erosion at registry scale and publishes its own failure rate.
The document carries a dated failure record as empirical input rather than confession: a federation declaration that advertised 342 fewer deposits than the archive held, closed before ratification and made uncomputable by hand; a functions-namespace collision that took 1,012 static files dark for six days while every local validator passed, because all of them read from disk; and a platform deployment limit that blocked publication entirely for twenty-four hours. Both of the latter are cited as evidence that single points of failure are observed rather than theorised, and that custody must be sequenced before ceremony.
A five-stage work plan follows, gated on a single condition the document states plainly: the entire roadmap activates when a second custodian exists. Stage I documents the mirror protocol that already works before specifying resolution on top of it, on the reasoning that one cannot specify distributed resolution before documenting how a peer fetches, verifies and serves. A nine-point acceptance test defines the finished system as a scriptable cascade, and eight invariants close the document as absolute constraints.