fix(doctor): keep the model's real basename in the gate self-test scratch dir

[MendixMau]

Direct lane (changes to skills/, bug-logs/, bin/, project-bin/, project-tests/, pipelines)

What changed and why (one paragraph):

bin/doctor.sh's gate_selftest() copied the project model into its scratch dir under a
hardcoded name — cp "$MPR" "$scratch/model.mpr" — and then copied mprcontents/ beside it
verbatim. mprcontents/ carries an internal record of the model's real basename (e.g.
App.mpr), so the renamed copy is a mismatched pair: mxbuild bails before writing any error
file, mxtk_mxbuild_error_count returns ?, and the self-test prints
gate self-test: gate cannot read mxbuild's error file (baseline run) with
GATE_SELFTEST_LINE="fail (unreadable error file)". That is the F-042-class string meaning
the gate itself is broken — so the symptom is a false FAIL on a perfectly healthy gate,
and it sends people hunting a problem that does not exist. The fix preserves the real basename
(scratch_mpr="$scratch/$(basename "$MPR")") and threads it through all four later references
to the scratch model: the baseline mxtk_mxbuild_error_count, the mxcli injection's
exec … -p, and the known-bad mxtk_mxbuild_error_count. scratch_mpr joins the function's
existing local list (bash 3.2-compatible, no namerefs). A comment at the copy site records
why the name must not change, so the next reader does not "tidy" it back.

Reported by Yvann during Mac/sandbox onboarding, 2026-09-22.

Field evidence — which real project this ran against / was observed in, and what it measured
or fixed there (CLAUDE.md → "Shipping an instrument", rule 4; required for any instrument change):

This has NOT been field-run, and the PR should not be merged as if it had. The authoring
environment is a Linux container with no .mpr, no mprcontents/, no mxbuild and no Studio
Pro, so the fix is reasoned from the code plus inspected, nothing more. The observation
itself is field evidence (Yvann, macOS, real project, 2026-09-22: healthy gate reporting
fail (unreadable error file)).

The run that would prove it: on a macOS machine with Studio Pro 11 and a real project,
bin/doctor.sh --gate-selftest <project-dir> must now print
baseline: gate read 0 error(s) off the scratch copy followed by the known-bad control
reading ≥ 1, instead of fail (unreadable error file). Reverting just the basename line on
the same machine should bring the false FAIL straight back — that revert is the proof, not
the pass alone. Please do not tell the team to pull until someone has done that.

• No client data anywhere in the diff
• Size cap: 2 files, ~20 lines changed (one of them CHANGELOG.md)
• Test tier reached: T0 inspection. tests/wave2/test-doctor-gate-selftest.sh is the
  only fixture covering this code. Its assertions are not greps over doctor.sh — it
  executes doctor.sh against a fake project ($P/Fixture.mpr + mprcontents/), a stub
  mxcli and five stub mxbuild behaviours. Read in full and checked against the change: the
  mxbuild stub takes the last non-flag argument as the model and looks for its
  .gate-selftest-bad-applied marker beside it, and the mxcli stub writes that marker
  beside whatever -p path it is given — both are basename-agnostic, so every case
  (healthy / no-file / always-clean / constant-nonzero / slow, plus the two-tree case)
  behaves identically before and after. bash -n bin/doctor.sh is clean. T2 is the
  outstanding tier, per the Field evidence box. Running the fixture would confirm no
  regression in the stubbed paths but cannot exercise the real-mxbuild/mprcontents
  interaction that is the actual bug — I have not run it; say the word and I will.
• For a new/changed instrument: golden input captured (not hand-written), both layouts
  (single-tree and .mpr-under-app/), both platforms (macOS + Git Bash/Windows) considered
  — both layouts are covered by the fixture's two-tree case and are unaffected (the change
  derives the name from $MPR, whatever resolved it); both platforms are unaffected
  (basename only). Golden input is the box above: not captured, and that is the gap.
• For a new skill: routing row added — n/a
• CHANGELOG.md line appended in this PR, crediting Yvann (CI-checked)
• New bug entries headed ## BUG-DRAFT-<slug>: — n/a, deliberately. bug-logs/mxcli-bugs.md
  is scoped by its own header to "Issues encountered … using mxcli + MDL. Collected for
  reporting to the mxcli team." This is a defect in this toolkit's own instrument, not in
  the mxcli CLI, so filing it there would send the mxcli team a bug they cannot act on. The
  CHANGELOG.md line is the record.

Related check performed (no change made, scope discipline): grepped bin/ and project-bin/
for other places that copy an .mpr to a scratch/temp location under a fixed name alongside
mprcontents/. None found — doctor.sh held the only $scratch/model.mpr, and
project-bin/snapshot-mpr.sh already does the right thing (cp "$f" "$DEST/$(basename "$f")"
plus mprcontents/ verbatim), with restore-mpr.sh as its mirror.

🤖 Generated with Claude Code

https://claude.ai/code/session_0187r9bQWkZAEppyEqrDevjK

---

Generated by Claude Code