metadata-protocol: a package-less row beside two packages' rows of one name (no shipper) is served by store order on the plain read while /layers refuses — follow-ups of PR #33182 (#32607) #33210

triage#1 · hotlong · session_01Q4EjjMnyLqJdzuktEmRPXH

Filing gate: ① a product defect with a named location, measured. These are the follow-ups of PR #33182 (#32607), named by review#1's PASS record 6870642184 (red-line-2 read). The PR does not move them; all are main's. Carried by epic:#31527#2 (6870658359). This is the one follow-up card for that PR (#32997 rule).

Checklist

• (a) A package-less row beside two packages' rows, with no shipper. That is three rows and no single holder. The plain read that names no package serves one of them by store order (rows.unbound), while /layers refuses 409 naming both packages. The registry already binds such a row to no package ("a customization of a name it cannot say whose").

  • Not a decision card: ADR-0150's red line 2, which the maintainer ruled on this epic, says ambiguity is reported, never guessed. Reading past the package-less row still leaves two holders, so both shapes end in the same answer.
  • The answer: the plain read refuses 409 RESOURCE_CONFLICT naming both packages, exactly as /layers does, and never serves the package-less row in their place. A row with a single holder stays readable bare.

• (c) A question, measure first. For a row-held base, the read naming no package at a dependent's added member serves the dependent's row (PR fix(metadata-protocol)!: a name two or more enabled packages hold only as stored rows, and no package ships, is refused by the read naming no package instead of served by save order (ADR-0150 D3) #33182's M2 pin). For a shipped base the same read answers 404 (metadata-protocol: an email-template locale added in a <base>.custom pair is read by name but never listed or sent, so a send in that locale renders the base's (#28779 follow-up) #28802). One operation should not answer two ways (SKILL.md: two implementations that disagree resolve to one), so align the row-held base with the shipped base's 404 unless the measurement shows a reason they differ. If it does, write the reason on this card instead of changing it.

• (d) Text fix in the ADR-0087 entry PR fix(metadata-protocol)!: a name two or more enabled packages hold only as stored rows, and no package ships, is refused by the read naming no package instead of served by save order (ADR-0150 D3) #33182 added (n3 of the merge-only delta record 6872223535, carried by epic:#31527#2, 6872234201). In packages/spec/src/migrations/entries/semantic/18.two-row-holders-bare-read-refused.ts, reason still says the read counts "the registry's holders and the packages its active stored rows are bound to". Since round 2, the registry's holders only help recognise a pair, as the changeset says. Make it say that. What the entry promises (409 naming each holder, and the ?package= remedy) is right and stays. This is a one-line text fix, and it rides whichever PR takes (a).

(b), one holder's layer chosen by save order, is folded into #32662 (the pair address after a reboot), as review#1 placed it. It is not tracked here.

Pins

Both save orders; a CONTROL with one holder plus a package-less row (unchanged); /layers and the plain read agree.

Reader

domain:server, under ADR-0150's red lines. The free epic seat epic:#31527#3 or the server lane may take it.