PATCH-001: be patch of a behind source drops a forward submodule-gitlink bump (checks out the sub, never stages the parent row)

be patch of a source whose ONLY forward change is a submodule gitlink bump checks the submodule out at the new pin but does NOT stage the parent gitlink row — so the merge silently leaves the parent pinning the OLD commit while the sub worktree sits at the NEW one, and be post records nothing. Hit while landing MEM-010 (an abc-submodule fix) per the dogfood flow. See Verbs, Submodules.

Issues

Observed landing MEM-010: worktree be://localhost/todo/MEM-010? was detached at 838e592b (behind main), its tip commit aa6c594a bumping ONLY the abc gitlink to f8bbaf19.

be patch 'be://localhost/todo/MEM-010?' on main printed sniff: patch: noop=239 take-theirs=0 merged=0 added=0 deleted=0 … (nothing absorbed — no patch row staged) for the parent, yet its submodule-recursion arm ran be get abc and checked the sub out at the new pin (upd CURL.c/CURL.h/test/CURLfw.c, checkout done).
Result is a half-applied, inconsistent state: the abc worktree is at f8bbaf19 (the fix is on disk) but main's committed tree still pins cd2057d0, and be status reports the parent clean — so a following be post commits nothing. The bump only landed after a manual be put abc + be post (a2e2300c).
Root hypothesis: the behind source's parent tree is — apart from the one gitlink row — an ancestor of main, so the parent weave classifies EVERY row (including the lone forward gitlink) as "theirs older → noop" and drops it; the independent submodule-recursion arm still advances the sub on disk. The forward-moved gitlink is the one row that must NOT be treated as behind.

Blockers

Adjacent to GET-017 (a detached/stale store-backed worktree won't FF in place, so be get refuses the dirty/cross-branch update — which is why the land has to go through be patch from the worktree-remote at all). PATCH-001 is the merge-side half: even via be patch, the gitlink bump is dropped.

Planned

A submodule whose pin moved forward relative to the target is an absorbable change regardless of the surrounding behind/ahead verdict.

In the parent weave, classify a gitlink row by comparing the two pins against their common base independently of the rest of the tree: a pin that advanced (target-base..source = forward) is take-theirs and must stage a parent put-row, so be post records the bump — matching the sub checkout the recursion already performs.
Keep the sub-checkout and the parent-gitlink-row consistent: either both advance or neither (no half-applied state); never leave be status reporting clean while the sub worktree sits ahead of the recorded pin.
Repro test: a behind worktree whose ONLY forward change is a sub gitlink bump; be patch 'be://…/<wt>?' then assert the parent gitlink put-row is staged and a plain be post records the bump (NO manual be put <sub> needed), and that the sub checkout matches the newly-recorded parent pin.

Outcome

Landed 5d5a8d2d (beagle, 2026-06-12): "stage forward sub gitlink-bump as take-theirs" — a forward-moved gitlink from a behind source stages the parent put-row, matching the sub checkout.
Later refined by PATCH-003 44df6916 (gitlink theirs-vs-BASE classification) and the 005..009 sub-recursion chain.