[Bug]: usage_statistics_environment tests fail when the checkout lives under a temp root (macOS /private/tmp worktrees) #8337

Preflight

• I searched existing issues and discussions for this behavior.
• This report is not an unpatched security vulnerability.
• The report and attachments contain no credentials or private/internal material.

Issue origin

Observed or reproduced in a real environment

LoopX version or commit

21013c0c1 (current main); also reproduced at 87a153d4d.

Host or runtime surface

macOS (any host whose checkout/worktree resolves under a temporary root, e.g. /tmp → /private/tmp). Not OS-version-specific; the same class would apply on Linux under /tmp checkouts where /tmp is on tmpfs.

LoopX area

Tests / usage statistics environment classification

Problem

Two tests in tests/control_plane_ts/usage_statistics_environment.test.ts build their "persistent" fixture with resolve(".local", ...), i.e. CWD-relative:

• "persistent machine routes retain notice-before-default-on without an enable flag" (:56) — mkdtemp(join(resolve(".local"), ...))
• "temporary roots, filesystem aliases and path-component boundaries are local eligibility facts" (:78) — persistent = resolve(".local", "usage-policy-persistent")

When the repository checkout (or a git worktree used for development) lives under a directory the classifier itself defines as a temporary root — on macOS /tmp resolves to /private/tmp, and /private/tmp is in the classifier's temporary prefix list (:100) — every "persistent" fixture path is also under a temporary root. The classifier then answers by its own correct rules:

not ok - persistent machine routes retain notice-before-default-on without an enable flag
  expected: 'notice_required'  actual: 'temporary_path'
not ok - temporary roots, filesystem aliases and path-component boundaries are local eligibility facts
  expected: false  actual: true   (the `configured + "-persistent"` case, :88)

The production classifier is behaving correctly; the tests assume the checkout is at a persistent location, which CI guarantees (/home/runner/work/...) but local contributor worktrees commonly do not (git worktree add /tmp/...).

Minimal reproduction

git worktree add /tmp/loopx-repro origin/main && cd /tmp/loopx-repro
npm ci --ignore-scripts
LOOPX_TEST_PYTHON=python3 PYTHONPATH=. node --no-warnings --experimental-sqlite \
  --experimental-strip-types --test tests/control_plane_ts/usage_statistics_environment.test.ts
# 2 fail — with the checkout moved to a persistent path, the same command passes 3/3

Verified on macOS arm64, Node 22.22.3: 2 failures from /private/tmp/... worktrees, 3/3 pass from a persistent-path worktree at the same commit.

Expected behavior

The suite passes regardless of where the checkout lives: the tests should anchor their persistent fixture to a location that is persistent by construction (not CWD-relative), or explicitly skip with a stated reason when the checkout itself resolves under a temporary root.

Actual behavior

Two deterministic failures whose cause is the checkout location, not the code under test — which misleads contributors into reporting false main regressions (it mislead me for an hour of triage today).

Sanitized diagnostics

+ actual - expected
+ 'temporary_path'
- 'notice_required'

Additional context

Related environment notes filed in case they help other contributors: root npm ci is required in fresh worktrees or ~12 test files fail whole-file with Cannot find package 'typescript'; browser scenarios spawning a Python backend need the package installed (pip install -e .) like CI. This report only claims the CWD/temp-root sensitivity, verified at two commits with both failing and passing controls.