POST AA b22: refrain ordering, VVN shape, and residual -m citation keys#
aa-b22-refrain-order-dv_ClaOp50Max_MMv1r0p0_2026m09d11_13h10Task list#
k |
s |
Task |
Notes |
|---|---|---|---|
k4 |
s1 |
Sweep the reversed-order refrain — 116 sites. WORDING DECIDED, TIMING DEFERRED. |
Language Rule 7’s refrain corollary requires the disbelief clause first and
the sentence to end on the task. The site does the opposite in 116 places
(measured 2026m09d11): 15 in DECIDED by LLoL 2026m09d11 — the replacement wording (see the table in section 1):
the single-sentence transform DECIDED — constraints. “Don’t touch the published papers.” Nothing in
When it is picked up, the 15 metadata sites are the separable first bite: they are not paper text, carry no version-bump argument, and are what a reader meets before arriving — the position where “not believed” does its most dismissive work. Precedent to weigh: Language Rule 5 grandfathered its own ordering rule (“applies moving forward; no need to sweep existing text”). Either the refrain corollary gets the same explicit grandfather clause, or the sweep gets scheduled — the one option to avoid is leaving the rule and the site silently disagreeing. |
k3 |
s1 |
Decide the VVN shape for the Matheo papers — three shapes are in use. |
Raised by LLoL 2026m09d11 on seeing
The third shape is redundant twice over: it repeats NOT changed. Editing one paper’s VVN splits it from its 47 siblings; editing all
48 is a site-wide decision, and LLoL’s ruling on the refrain — “now isn’t time for
such evaluation work” — applies here with equal force. Where it is used:
exactly one place per page, the |
k3 |
s1 |
Resolve the 7 remaining unresolved |
After the 2026m09d11 bib repair, site-wide empty |
1. The refrain: why this is a real decision and not a typo sweep#
The rule postdates the text. The refrain corollary entered CLAUDE.md on
2026m07d14; the MMv5 floor was poured on 2026m05d29. Every one of the 111 sites was
written correctly by the standard in force at the time. This is inherited debt, not a
violation — and the distinction matters, because Language Rule 5 sets an explicit
precedent for exactly this situation: the BREACH-before-HELD ordering “applies moving
forward; no need to sweep existing text across the site.”
So the honest question is whether the refrain deserves a stronger treatment than Rule 5 gave its own ordering rule. Three things argue that it might:
It is the project’s most-repeated sentence, not an incidental phrase. 111 sites is close to saturation of the public surface.
It sits in
og:card:descriptionfields, so it is what a reader sees before they arrive — the position where “not believed” does its most dismissive work, and where the rule’s stated reason bites hardest.The four correct-order uses that do exist are on newer pages, so the site is already drifting into two standards. A split standard is worse than either standard alone.
And three that argue against sweeping now:
111 edits across 39 published papers is a large content change requiring version bumps on each, for a wording change that alters no claim.
Rule 5’s precedent says moving-forward-only is an acceptable answer for ordering rules.
"critiqued, not believed"may by now be a recognised tagline with its own recognition value; changing it site-wide has a cost the rule did not weigh.
DECIDED 2026m09d11 — wording yes, sweep not yet. LLoL approved the single-sentence transform and rejected the two-sentence variant that had been offered for the 66x form (“the 2-sentence variant is longer and both belong in one sentence”), and ruled that the sweep itself waits: “Leave it as is for now. Maybe future work will find a better slogan, but now isn’t time for such evaluation work.” Binding constraint: “Don’t touch the published papers.”
So the replacement table above is settled wording, parked work. Whoever picks this up should not re-open the wording question; they should decide only when, and whether the corollary instead deserves an explicit grandfather clause like Language Rule 5’s.
The one option still to avoid is silence — a rule that 116 public sites contradict is in an unstable state, and leaving it unexamined is the form-looks-right-but-substance-drifted pattern the Core Principle names as the most dangerous because it is invisible. Parking it here, with the wording pre-agreed, is what keeps it from being silence.
When it is picked up, the 15 metadata sites are the separable first bite:
og:card:description, :description: and :keywords: are not paper text, carry no
version-bump argument, and are what a reader meets before arriving — the position where
“not believed” does its most dismissive work.