Motir

Found by motir run MOTIR-3878 while executing MOTIR-4206 (the public-tenant-addresses ADR), 2026-09-03.

What the card asserted

MOTIR-4206's Q7 row offered a rung-1 figure as the grounds for its recommendation:

Renames are capped (Atlassian caps at 3 — decide a number).

and listed the source in its own Context refs:

Atlassian — https://support.atlassian.com/organization-administration/docs/update-your-product-and-site-url/

What the cited surface actually says

Fetched 2026-09-03. The URL in the card's refs carries no cap at all; the figure lives one page over, at https://support.atlassian.com/organization-administration/docs/update-a-product-url/, and it is:

"You can update the subdomain up to 15 times."

So the number is 15, not 3 — a factor of five, in the direction that makes our own cap look generous when it is in fact stricter than the mirror's.

Two smaller readings from the same sweep, recorded because they are the same shape and were checked at the same time:

• The card's Q7 also asserts Atlassian says a changed site URL "remains as a redirect" and cannot be re-used by another organization. The cited page states the redirect ("We'll store the previous URL in your site's history and redirect all links to the new URL") but not the no-reuse half; that half appears only in Atlassian community material. The ADR takes the no-reuse decision on ProjectKeyAlias's in-repo precedent (rung 2) instead, so it does not rest on the unverified reading.
• The card's framing of Statuspage as "the root domain is explicitly not usable: it must be a subdomain" is not on the cited page either — as fetched it is a recommendation ("we recommend that you make this a dedicated domain"), not a restriction. The ADR accepts apex domains and records the difference.

Why this is a planning bug rather than a note

A rung-1 citation is the strongest kind of grounds a card can offer, and its whole value is that the reader does not have to re-fetch it. A wrong one is worse than an absent one: it is inherited silently by every card that reads the record, and the ADR's Q7 recommendation was built on it. The runner caught it only because the card's own acceptance criteria demand the observation be quoted as fetched — a card without that criterion would have copied the 3 straight through.

The general shape: a planning pass wrote a specific number from memory into a slot the card itself marked as needing a fetch. The number was plausible, the URL was real, and the two did not correspond.

What was already done on this run

• The ADR records the correction in full — docs/decisions/public-tenant-addresses.md §8 carries the fetched figure, the quote and the URL, and decides 5 with the deviation's use case stated (a rename permanently burns a name in a shared namespace; our namespace is new and small where atlassian.net's is not).
• MOTIR-4206 is amended on the record — its Q7 row now carries the correction inline rather than the superseded figure.
• The ADR's §14 Sources carries a "Read but NOT confirmed" section naming the two secondary readings above, so nobody re-cites them from the record either.

Acceptance criteria

• No card in MOTIR-3878's subtree still asserts a rename cap of 3 (grep the subtree's bodies for caps at 3).
• The takeaway is captured as a lesson (done on this run — see the linked card's comments) or the reason it was not is stated here.

Context refs

• MOTIR-4206 — the card, amended
• motir-core/docs/decisions/public-tenant-addresses.md §8 and its Sources section — the corrected record
• https://support.atlassian.com/organization-administration/docs/update-a-product-url/ — where the 15 actually is
• https://support.atlassian.com/organization-administration/docs/update-your-product-and-site-url/ — the page the card cited, which carries no cap