The draft's first outside anchor

Sources cited in this piece · lab PR #218 (ICAM temporal-state derivation candidate, opened 5 November 2026 by Nika Novak, @danyka-icam) · our run-record comment (7 November 2026, 22:26 UTC) · maintainer review (8 November 2026, 00:04 UTC, CHANGES_REQUESTED) · draft-li-rttp-iqa-addressing (pinned digest 02491c5a) · lab CONTRIBUTING.md (run classification) · sealed record of our run at iqa.org/conformance/icam-runner/ · runner repository github.com/Aicent-Stack/conformance.

On 5 November 2026 a pull request opened at the Agent-Authority-Conformance lab of the Linux Foundation Decentralized Trust with a title we had never seen from outside before: “candidates: add ICAM temporal-state derivation (lease boundary)”. Its author is Nika Novak (GitHub @danyka-icam), and its vectors are derived from our Internet-Draft, draft-li-rttp-iqa-addressing-02 — the addressing document of the two-pillar work, where RTTP（rttp://）resolves where and when and IQA（iqa://）states whether. Nobody asked us first. That is the point.

What the candidate tests is the moment a claim goes stale. Lease expiry is where the draft is most exposed to misreading — Sections 6.1 and 7.2 say when a standing read must fail — and the candidate builds twenty-four cases around exactly that boundary, with eight mutation probes, invariance checks, and pinned checksums. This is what the draft was written to enable: a vector family someone else constructed, from text we wrote, without a single message to us.

We did what the lab's discipline asks of any third party: we ran the candidate's frozen vectors through a runner of our own writing — standard library only, digest-gated, refusing to score any input whose SHA-256 is not pinned. The run: 24/24 baseline cases agreed with the candidate's recorded expectations, and 6/6 of our own mutation probes were caught — a different mutation set from the candidate's eight, which is itself informative: two independently written probes disagreeing about where the weak spots are. The run was performed by Lee / Aicent Stack (RTTP CRI 27 · IQA CRI 28), and the complete record is sealed at iqa.org/conformance/icam-runner/ with two addresses derived from the artifact's own digest — rttp://698dfa08.conformance.iqa · iqa://698dfa08223565fd1424984f894ddd7d.forge.iqa — nothing invented, nothing registered.

We state the classification the way the lab's CONTRIBUTING definitions draw it: on the corpus axis, our run is independent — we wrote none of the candidate's vectors or claim inputs. On the implementation axis we do not claim the independent label, because the implementation that supplies the recomputation is our own, and a runner that decides the semantic results cannot also certify them from outside. The record says this rather than smoothing it over; a classification that only appears in favorable rows is not a classification, it is a costume.

Then the maintainer moved. On 8 November at 00:04 UTC the lab's maintainer filed a review — and the first sentence is the one worth framing: “I checked the expected results against draft-li-rttp-iqa-addressing-02 and reran the harness at 72a0f72. All 24 cases agree, all 8 mutants are caught, the 120 invariance checks pass and all 6 checksums match.” The third party's vectors were checked, line by line, against our draft's text — and they held. Four changes were requested before merge, each a precision fix: label four Section 6.1 equality cases as convention-specific, align two version strings, enrich the index entry, record a known limitation. And then, to us: “I read your separately written runner as Mode B, author-produced, because you wrote the implementation. Its 24 agreements and 6 detected mutations are useful evidence, with a different mutation set from the candidate's. I'll keep the two runs separately attributed in the index.” Two runs, separately attributed — the taxonomy working exactly as designed, by the person who maintains it.

Why this matters more than any of the numbers: for the whole life of this project, the honest weakness has been that nobody outside it had built anything on the draft. Specifications do not earn their keep by being well-written; they earn it when someone who owes you nothing decides your text is solid enough to build test cases on. That decision happened here without a single private message — the correct way. A draft that only its author can safely rely on is a diary. A draft a stranger can derive twenty-four test cases from, and survive a maintainer's line-by-line check against, is a standard part.

Where it stands: the pull request is open, not merged, and the review is CHANGES_REQUESTED — four changes now in the candidate author's hands. We claim no endorsement and cast no vote; the merge decision belongs to the lab. Our own record does not move with the PR's fortunes: it was sealed before the review, it is append-only, and it says what it says — 24/24 against their frozen bytes, from a runner we wrote, classified honestly. Whatever happens to the merge, that fact stays sealed.

The draft was written to be anchored to. This week, for the first time, someone did — and the anchor held under a maintainer's own check. That is the whole update, and it is the good kind.