Twinned from https://github.com/cfsh/ninefold/issues/745 by tools/beads/import_github.py's reconcile pass.
Suggestions from ninefold-reviewer's review of #739. All four are sub-floor — they were carried across three verdict rounds without ever gating, and #739 converged with them open. Unclaimed, unlabelled; take them or close them.
R1 · a corrected sentence left behind in the test that pins it. tools/tests/test_probes.py, ImportCacheTest.test_an_UNKNOWN_probe_NAME_refuses_BEFORE_paying_for_an_import — the docstring still says the drifted name *"must refuse in milliseconds rather than after an engine launch"*. #739 corrected exactly that wording in probe_all.py (to *"without an engine launch"*, because scripts() still runs after build_once, and #380's build-first rule is what stands in the way of moving it earlier). The copy in the test asserting the ordering was not.
R2 · the force failure message can name the wrong import. tools/nflib/godot_import.py, ensure():
``` absent = (f"; {len(after.absent)} artifacts absent" if after.absent else "; every artifact is present, but from an EARLIER import") ```
The else-branch is reachable on the force=False arm, where before.absent was non-empty by construction — so anything present afterwards arrived from *this* run, not an earlier one. Diagnostic-only, and a narrow arm (import exits non-zero yet still writes every artifact), but the sentence is the opposite of what happened. Gating the wording on force would settle it.
R3 · trouble is now multi-line and both callers format it as one line. Since the D1 fix, the reason string carries a tail-capped block of engine output. probe_all.sweep prints err(f" {trouble}."), so only the first line is indented and the full stop lands after the engine tail; probe.probe has the same shape. Cosmetic — probe_gate.parse_sweep matches the marker line and is unaffected.
R4 · doctor() walks the tree once per call. presubmit.py's import-cache check calls godot_import.scan(...), ~42 ms, and tools/tests/test_presubmit.py calls doctor() about ten times: ~0.4 s on a ~7 s presubmit. Halved by #739's one-walk Cache fix, not removed. The jcantsp-bot reply on #739 recorded this as *"left"* deliberately — a tree walk in a diagnostic — so this line exists only so it is not re-derived a fourth time.
None.
No comments.