`CLAUDE.md` tells every session that ADRs are the one record family still

CLAUDE.md tells every session that ADRs are the one record family still hand-numbered. They are not, and have not been since decide adopted the mint.

The claim

Under Concurrent sessions, on record ids:

ADRs keep their hand-numbered filename ordinal, so an ADR is the one record family where minting from two checkouts still needs a word first.

The reality

core/decide mints an adr-<stamp> through core/recordid like every other family. internal/README.md:104-112 records the adoption and its date: decide "is the last record family to reach that seam (the 2026-09-01 ruling, the turn adr-45 ruling 3 deferred), and it adopts it the way every family before it did — by holding a recordid.Minter and naming its family tag, never by carrying an allocator of its own."

internal/core/decide/decide.go:48 holds that recordid.Minter. Nine ADRs in .abcd/development/decisions/adrs/ already carry minted ids, the newest being 2609091248200336-a-tool-never-creates-directories-in-user.md — which CLAUDE.md itself cites, two sections above the claim that says such an id cannot exist.

Why this is major rather than cosmetic

It is not prose in a chapter somebody may read. It is a rule in the file the rules loader puts in front of every session, and it tells an agent that a coordination step is required where none is. The predicted failures are ordinary: an agent pauses to ask before minting an ADR, or hand-numbers one to obey the rule and lands a 00NN- filename beside the minted ones, deepening the very split the mint was adopted to end.

The sequential ids that remain are not counter-evidence. Minting is forward-only and nothing renumbers an existing record, because an id is a citation: 50 sequential ADRs and 9 minted ones is what a family mid-adoption looks like, not a family that hand-numbers.

Scope of the correction

The rule's surrounding sentence is right and should stay: ids need no coordination between checkouts because the allocator reads no maximum (adr-45). What must go is the ADR exception carved out of it. Check the same claim wherever else it was copied — .abcd/development/brief/ and the decide surface chapter are the likely carriers, since the exception was true when they were written.

Acceptance

• Given a session reading the rules, when it reaches the record-id rule, then no family is named as an exception to minting.
• Given the corpus, when the claim is checked, then no committed surface states that ADRs are hand-numbered.

Grounds

• pursued: the router is the surface an agent reads first, so the correction holds only while the test that mints and then reads the bullet stays armed; it would be shown wrong if a rewrite restated the exception in words the containment check does not carry