[MendixMau]

Supersedes #593 by @saingge, with the maintainer's decision on that PR (option 2, narrower). saingge is the commit author.

Direct lane

What changed and why: after the write (MDL and --patch) and before the mxbuild gate, exec.sh runs model-scope.py restore (already on master). Units that changed only in XPath whitespace go back to HEAD with their ContentsHash rows updated, so the gate checks the final model. A scope: note goes in the BUILD-LOG row. The step never blocks: a failure is a warning, MXTK_NO_SCOPE=1 skips it, and it is skipped when FORCE_EXEC let the write run over uncommitted model changes.

Dropped from #593, by decision: the pre-commit hook refusal (marker v2), --check reporting v1 hooks as stale, and the hook tests (fixture section E). model-scope.py check --staged stays available to run by hand. So no synced project gets a new commit-time refusal.

Research: is it still needed on mxcli v0.29? I re-ran the #1596 rename on the real public binaries, on the fixture model with the golden capture's steps, and classified the result with master's model-scope.py:

mxcli	units changed	real	noise
v0.28.0	3	1	2
v0.29.0	1	1	0 (fixed upstream, ako/mxcli#1506)

So the step is insurance for projects on mxcli ≤ v0.28 and a no-op on v0.29.

Row	Answer
Process	none new; nothing blocks
Dependencies	project-bin/exec.sh: one step after the write and before the gate, plus ${SCOPE_NOTE} in the BUILD-LOG cell. The conflict was resolved onto master's current BUILD-LOG line (timestamp fast path kept), and #603's hash verify is untouched
Context cost	none
Behaviour	n/a: script only
Runtime	one model-scope.py restore per exec on a committed model (one Python start plus a git diff). Not measured on Windows: a reviewer item

Re-verified on master 0db1c0e: test-model-scope.sh passes 50/1 here and 43/8 against master's exec.sh (section F: 7 new failures). The single failure on both is section C's partial-restore case: chmod 0555 cannot stop root, and this container runs as root. It fails identically on master's own fixture here, so it is an environment artifact, not this change. CI runs it as non-root.

• No client data
• Size: 4 files, +127/−3
• CHANGELOG line rewritten for the reduced scope (credited to saingge)

Review: pending independent review

---

Generated by Claude Code