AICENT.COM
← Back to Blog

Labels don't replay — a note on moderation and verifiable work

2026-11-18 · IQA · RTTP
Sources cited in this piece · the text under discussion: draft-li-rttp-iqa-addressing (the exchange itself lives in a public IETF list archive, retrievable by the subject line of the material it shut down) · the family on the table: iqa.org/conformance/aicent-grant-ref/ · recorded in a third-party lab mainline: candidates/aicent-grant-ref · the conformance hub: iqa.org/conformance/ · the registrations: rttp, CRI 27 · iqa, CRI 28 · Sister pillars: RTTP (rttp://) · IQA (iqa://)

This week, on a public IETF working-group list, an eight-post technical exchange about the executable conformance material of 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 — was shut down by the list's moderator. The shutdown post was three sentences long: the work "is not adopted"; the messages are "clearly Claude loops"; and the discussion should move off-list and come back "when there are concrete WG actions such as a review or adoption call", closing with "Don't make me lock the list again!". No names appear in this piece, and none are needed: every timestamp below is public and checkable by anyone holding the subject line of the material that was shut down. This piece does not defend, does not appeal, and does not argue. It does the one thing the shutdown post did not do: it shows its work.

Two clocks, one archivePRIVATE CLOCK — 13 NOV: moderation query to the chairspost location · message-id · a choice offered: release it, or discard it for a clean resendsilence — four days and counting · no reply on the recordPUBLIC CLOCK — 17 NOV 16:00 UTC: our last post of the threadeighth of eight · a technical answer, filed and archived like the seven before it17 NOV 21:11 UTC — the shutdown post · five hours later04 NOV — the original broadcast held by the spam filter · never released to this dayprivate obligation: four days of silence · public gesture: five hours · every timestamp publicthe exchange is retrievable from the public archive by the subject line of the material it shut down

The clock nobody published

Start with the record. On 4 November we posted executable conformance material for the refusal vocabulary and lease semantics of the draft — a closed five-value refusal set, a fixed priority order, grant references, and two vector families with recorded, replayable runs. The list's spam filter held the post: our mail relay had encoded the subject line, and the list's header rules quarantined anything encoded. It has never been released. On 13 November we sent the chairs a private moderation query — the post's date, its full subject, its message-id, and a plain choice: release it, or discard it so we could resend cleanly. Four days later, there is no reply on the record. On 15 November we resent the material with a plain-ASCII subject; it went up, and what followed was three days of genuinely technical exchange — eight posts, four from each side, spanning refusal vocabularies, layer boundaries, and a third-party recomputation of our vector family that has since merged into a lab mainline. Our eighth post went up at 16:00 UTC on 17 November. At 21:11 UTC the same day, the shutdown arrived. The arithmetic is the whole critique: the private moderation obligation sat in silence for four days; the public gesture took five hours.

The rule that is not written

"Come back when there are concrete WG actions such as a review or adoption call" has the cadence of a procedure. It is not one. Nothing in the IETF's written processes requires that a draft be adopted, or even scheduled for review, before its material may be discussed on a working-group list; individual submissions are discussed on WG lists every week — reviews, test reports, first-implementation notes — and the very list in question had hosted exactly such threads in the months before this one, without anyone demanding prior WG action. A chair is free to prefer a quieter list; a chair's preference is not a rule. The distinction matters because rules can be cited and applied evenly, while preferences get applied to whichever thread has lost the room's attention that week — and "clearly" is the tell: the post asserts without showing, which is precisely the failure mode our own material is built to make impossible.

On the label itself, one structural observation and one structural answer. The observation: "clearly Claude loops" was aimed at both sides of the thread — it swept up an independent implementer whose recomputation of our family now sits in a third-party lab's mainline, merged with separate attribution. Whatever the label was meant to price, it mispriced that. The answer: the label does not matter, because of what the thread was actually about. The material under discussion consists of vectors frozen behind digest gates, recorded outputs that replay byte for byte, sealed artifacts each carrying two addresses, scheme numbers registered at IANA, and lab mainline entries attributed to named operators. None of that changes depending on who typed the words — a human, an assistant, or both together — because the claims are not in the prose; they are in the bytes, and the bytes replay. A label is an assertion with no digest: it cannot be rerun, cannot be diffed, cannot be cited in a year except as a mood. That is why we will not defend against it. Defense accepts the frame, and the frame is weaker than the artifacts.

A label vs. a recordTHE LABELan assertion · no digest · no inputs namedcannot be rerun · cannot be diffedcannot be cited in a year except as a moodexpires with the room's attention · costs nothing to writeTHE RECORDvectors frozen behind digest gates · inputs namedoutputs replay byte for byte · sealed, two addresses eachschemes registered at IANA · operators namedthird-party lab mainline entries · separately attributedthe way the words were produced cannot tilt a replayable score — the claims are in the bytes,and the bytes replay regardless of who typed the promptdefense accepts the frame · the frame is weaker than the artifacts · we decline the framethe only reply a label cannot answer is a replay anyone can run

Walls die; artifacts live

The double standard — private obligations slow, public gestures fast — is not a quirk of one moderator; it is a structural incentive of moderated walls. A wall's product is attention, a moderator's currency is the room's mood, and the cheapest way to display authority is a shutdown post: it costs nothing to write, cites nothing, and cannot be falsified because it asserts nothing checkable. Our records age differently. The thread that was shut down contains a merged third-party recomputation of our vector family; the material it dismissed is pinned in a lab mainline, bound to registered scheme numbers, and sealed for replay. When the wall forgets, the archive does not; when the room's mood moves on, the digests stay where they are.

Our actions, stated plainly so nobody has to infer them: nothing further on that wall — no appeal, no defense, no return until "concrete WG actions" exist, which is a bar this material was never aiming at, because it moves through the individual-submission and editor channel, where review is a schedule rather than a mood. We have since closed our subscription to that list — not as protest, as plumbing. The work answers the label in the only language a label cannot speak: the revision continues on its own schedule, the vectors stay frozen, and every claim in this piece is checkable from the sources above. The room can keep its labels. We will keep the replays.

One line: a shutdown post expires with the room's attention; a replayable record does not — rttp resolves where & when; iqa states whether — and where the two agree, action begins.

SEALED rttp://f6aa5bbf.blog.aicent · iqa://f6aa5bbfb148dd919e0b0861938c57e1.forge.iqa · seal record