Whose queue is this? The seat’s (2026-08-28). The PO’s three-role model answered the IA question the next paragraph was asking: an owner’s queue IS on the admin surface, scoped to that owner’s applications and nothing else, and the Super Admin sees every owner’s in one list. Two things followed. With other owners — the read-only view of other owners’ pending requests, marked as a tension since it was drawn — is gone, because “sees only data related to their apps” rules it out; the strip that selected it now filters by application instead. And the Super Admin can decide a request on an owner’s behalf, which puts a central operator’s name on a business-access approval the operating model gives to the owner (§13) — drawn because the PO said “can do anything”, and marked beside the buttons. A request with no owner stays undecidable for everyone, including the Super Admin: the fix for a missing owner is naming one, not standing in.

As it stood until 2026-08-28 — whose queue is this? Su Su Example's, as the owner of PDO Finance and Operation Solid Rock — not the identity administrator's, and never Engineering's (§13). It is drawn inside the admin shell for prototype navigation. The second section, With other owners, exists so the whole portfolio's picture is visible somewhere; but nothing in the requirements gives one owner sight of another's queue, so whether this page belongs on an owner's surface, in the portal, or only on Central Operations' is the open IA question this screen exists to surface. The persona switcher does not change it: the seat is fixed, because the admin surface has never read the persona and the default persona owns nothing.

Why the panel has a scope selector (2026-08-27). Measured on the Busy preset at 1440×900, the three stacked sections split the panel’s scroll length 35% / 12% / 52% — so the majority of it was With other owners, the one section with nothing to press and the one its own marker calls a tension. Every row below the fold was something an owner cannot act on. Tabbing the three would have made them peers, which promotes that disclosure from buried below to a normal mode of this screen — more commitment, not less. The selector instead moves along the axis the 52% actually sits on: whose decision is this. It re-orders nothing, so oldest-first and the note at the foot of this page are both untouched.

And then it became a partition rather than a widening, on the designer’s call the same day: Others · 5, not Everyone · 10. Selecting it takes your own queue off the screen, which the superset never did — and the count is the better one, because the question an owner has is how many are not mine and the superset made them subtract to get it. Nothing is lost by not seeing both at once: they were never interleaved. ⚠ It also settles the tab style. The pill = pick a peer, line = switch a panel rule made the underline a violation while this was a widening, and makes it correct now that the two positions swap the panel’s contents.

What the headings cost on the way out. Both repeated the tab directly above them, so they went. But Yours to decide was also the list’s accessible name (now an aria-label) and both were carrying markers, which moved to the strip under the tab bar. 🚨 The per-owner headings went too, and that one is a real loss rather than a tidy-up: from the list alone an owner can no longer tell whose decision a request is, and the Green Light request nobody owns no longer looks different from the rest. The detail pane still says both in full.

The Quiet preset is the pilot's truth — 20–30 people, about one request a week — and the Busy preset is the 100-user wave. Triage beyond oldest-first, and pagination, are deliberately not designed: whether the queue needs them is a finding worth having with numbers, before the second wave rather than after.