Beagle SCM

JGET-014 LOW: same-line weave — jab escalates to cnf/GETCONF (rc=1) where native emits mrg (rc=0), and ours/theirs are SWAPPED

A behind-worktree FF / cross-branch weave over a file with DISJOINT edits merges correctly under jab (verified PARITY). But a directly-COLLIDING single-line edit diverges twice from native: jab emits a fatal cnf + GETCONF (rc=1) where native emits a non-fatal mrg (rc=0) with the union markers, AND the conflict hunk has ours/theirs reversed (<<<<theirs||||mine>>>> vs native <<<<mine||||theirs>>>>). Part of JGET-001; method Issues.

Input

Context

• Native beagle/test/get/66-staged-overlap-merge (arm B requires mrg, rc=0).
• Repro: staged/unstamped overlap with a same-line collision; jab → cnf+GETCONF

  rc=1, sides swapped; native → mrg rc=0. Disjoint-region edits: PARITY (verified). Evidence /tmp/claude-1000/jabprobe (case 66 pair).

• Root: the 3-way weave (get.js:748 / shared weave3) — conflict classification

  (fatal vs non-fatal mrg) and the ours/theirs argument order.

Goals

• A same-line overlap weaves to mrg (rc=0) with the union, ours/theirs in native

  order; only an unmergeable case escalates — matching native case-66 arm B.

Constraints

• Match native's conflict taxonomy + marker side order. Repro-first.

WIP

Design decisions

• Audit weave3's conflict path: confirm the mine/theirs argument order vs native,

  and the mrg (emit union, rc=0) vs cnf (GETCONF, rc=1) decision boundary.

TODOs

• PREMISE RULING FIRST: the cnf/rc=1 escalation is now PINNED as spec by be/test/get/dirty case A2 (d80bdb5c, TEST-003 — native retired as oracle, "a real conflict must exit NON-ZERO" + markers required); native's mrg/rc=0 is no longer the target unless DIS-057/Dirty.mkd rule otherwise. Decide, then either wontfix this half or change the test.
• The ours/theirs ORDER half is still live: leaf calls weave3(baseBytes, onDisk, bytes, …) (get.js:774) with ours=onDisk (dirty), theirs=target; wm.merged([oScope, tScope]) (get.js:904-907) renders ours-scope first — verify the fence layout against the reported <<<<theirs||||mine>>>> swap and fix the side order if confirmed (dirty A2 asserts MINE5 presence, not position).
• Spec gate: extend be/test/get/dirty A2 to assert the marker SIDE ORDER.

Blockers and bummers

• Overlaps DIS-057 (be/ dirtiness/merge classifier vs Dirty.mkd) — coordinate the

  mrg/cnf taxonomy there.

• AUDIT 2026-07-07: half the ticket (escalation) is likely moot post-TEST-003; left open for the order half + the ruling.

Outcome

• (open) Filed 2026-06-29.