Three forms of adoption

Sources cited in this piece · draft-li-rttp-iqa-addressing-02 (pinned TXT digest 02491c5a, sections 6.1, 7.2, 8.1, 8.2) · registry entry predicate/v1/refusal-reason-priority (live, verified 11 November 2026) · registry PR #544 (merged 10 November 2026, commit f76e5fe8) · corpus PR #483 (merged 7 November 2026) · author confirmation in #492 · lab PR #218 (merged 8 November 2026) · WIMSE list posts by Cliff Singletary and Jiahui Miao (6 November 2026) · our conformance pages iqa.org/conformance/ · the lab Agent-Authority-Conformance/aps-conformance-suite.

There is a difference between being cited and being adopted. A citation quotes you; adoption builds on you — someone else's system takes a piece of yours and gives it a load-bearing role of its own. This post is the ledger of one such week for the two-pillar work, where RTTP（rttp://）resolves where and when, and IQA（iqa://）states whether: an independent registry registered our closed refusal vocabulary as a formal type, a third-party conformance candidate anchored its cases to our pinned draft, and two participants on the WIMSE list adopted our split in their own words. Three forms — and one boundary we will not blur.

What got registered, exactly. Our draft's Section 8.1 defines the refusal reason as a closed set of five values — evidence_binding_mismatch, evidence_expired, authority_not_determined, authority_revoked, operation_not_permitted — and the reason must be the first value in the list whose condition applies: the priority order is part of the definition. The agent-evidence-vectors project — an independent evidence-registry effort run by Gilda (astrogilda) — merged our corpus into their mainline on 7 November (#483) and, three days later, registered the vocabulary as a formal predicate type (#544): the type URI is live, the registry's RFC table points its rttp-refusal slot at it, an example record in their registry matches the entry by digest (8fdf1041), and their corpus checker answers 33 of 33 cases. The entry states, in the registry's own words: "A later revision that keeps the set and the order is a new confirmation; any change to either is a new URI." Our closed set — and its order — are now a managed asset in someone else's system.

Adoption invites a harder question than citation does: when someone builds on your work, do they attribute it? So we audited our own adoption the way we would audit anyone's — four gates, run against the merged mainline rather than the pull-request descriptions. Gate one: the registry's manifest names the provenance in its own files — author, source thread, pinned draft revision. Gate two: the signature pins are present where we would put them. Gate three: the sha the registry cites for the draft (02491c5a) compared byte-for-byte against our own archived copy of the same file — they match. Gate four: derived work says derived; the registry's README asks its members to write derived from when they go beyond the author's records, and they do. The audit script is fixed and re-runnable; the result is 4 for 4, on the record.

The second form sits in the conformance lab. PR #218 by Nika Novak (danyka-icam), merged on 8 November, is the first third-party candidate in the lab whose cases derive from our pinned draft rather than from any oracle we wrote: twenty-four cases, each traceable to the text, with the draft's digest pinned in the candidate document. That is a different kind of reliance than the registry's — not our vocabulary as a type, but our text as the source of someone else's test logic. Our own interop record against the merged candidate is in the queue as #225, and the family pages it exercises — aicent-grant-ref and aicent-lease-boundary — remain downloadable with their checksums, with the lab's queue as the public counterparty for anyone who wants to rerun them.

The third form is made of words, and we count it honestly. On 6 November, on the WIMSE list, in the thread on draft-ietf-wimse-aims, Cliff Singletary restated the boundary our closed set draws — a determination outside the set's conditions is "recorded as not determined, not as not permitted" — and Jiahui Miao offered the sentence we have since credited and will carry into our next revision: "a determination made from observed evidence holds at the time of observation; it does not by itself hold after that evidence expires." Neither post links to our artifacts; this tier is agreement in public, not integration. It belongs on the ladder because words on a standards list are how artifacts start — but it stays on the bottom rung, and this piece does not move it up.

And the boundary. What exists, as of this week: normative adoption in three forms — a registry type, a derived candidate, a named split. What does not exist, and this piece will not pretend otherwise: a production implementation of the schemes, and interop between two independent implementations. Those are the next rungs above the ladder, and nothing on this page shortens the distance to them. The draft does the promising. The adopters, this week, started keeping the receipts for us.