main red: audit-live-rows reports the Music3 ACTIVE row abandoned, so preflight is red on a clean main

[localai-bot]

What

On a clean checkout of origin/main (22367c551), two gates are red:

$ python3 scripts/audit-live-rows.py --check    ; rc=1
$ python3 tests/scripts/test_audit_live_rows.py
FAIL: test_shipped_record_has_no_abandoned_active_row
AssertionError: stale ACTIVE rows remain:
  ['MODEL-MUSIC-minimax-music3-mini-max-music3-for-conditional-generation']

The auditor reports the row as ABANDONED: "no branch, no commit on main
mentioning the row ID".

Why

7a0e6c82b (#723) moved the Music3 row to ACTIVE, but the auditor's liveness
rule wants the row ID named by a commit on main or by a live branch, and the
merged squash subject does not satisfy it. So agent-preflight.sh is red on
main itself and on every branch cut from it.

Reproduced in a detached worktree at origin/main, so this is not a
working-tree artifact. Found while landing #699.

Fix

Either the row's record names its landing commit the way the auditor expects, or
the liveness rule is the thing that is wrong and gets fixed with its own
red-before evidence. Not by relaxing the assertion.

[localai-bot]

Self-resolved, and the mechanism is worth recording rather than just closing. The auditor's liveness rule wants the row ID named by a commit on main; 25df7468f (feat(MODEL-MUSIC-MUSIC3): W4/W5 …) now does, so the row reads LANDED instead of ABANDONED. On main 6300fb8ab: audit-live-rows.py --check returns rc=0 on three consecutive runs, 0 abandoned ACTIVE, and test_audit_live_rows is 41/41.

Note for anyone who hits this again: the verdict flips with main's commit history, not with the tree, so the same checkout can report rc=1 and later rc=0 with no edit in between. I saw exactly that while landing #699 — it is not flakiness in the checker, it is the rule being time-dependent by design.