Remediate 19 scripts/ drivers with repo_root()-derived sys.path inserts, then widen the #2181 check to scripts/

Goal

Fix the 19 scripts/ drivers that derive a sys.path entry from task_workflow.repo_root() via a one-hop module constant, then extend workflow_lint.py's check_no_repo_root_syspath_in_tests to cover scripts/ (renaming it accordingly).

Why

repo_root() branch-guards to the MAIN checkout. A driver that does PROJECT_ROOT = repo_root() + sys.path.insert(0, str(PROJECT_ROOT / "scripts")) therefore imports main's copy of its sibling modules when run from a worktree — so a branch-local fix to a sibling module is silently not exercised, and a foreign checkout's scripts/ is leaked onto sys.path for the process.

This is the same latent trap #2164 hit on the tests/ side, where it cost ~10 probe commands plus a ~20-minute gate re-run to diagnose (the leaked path defeated the #1296 negative control in tests/test_backend_poll.py). The tests/ half is now permanently blocked by the #2181 check; this task closes the scripts/ half.

The 19 offenders

Enumerated by two independent AST sweeps at #2181 plan time (the plan critic's and the code reviewer's, concordant), each re-implementing the predicate and running it over the live tree — scripts/: 19 flagged, tests/: 0, src/: 0.

17 × scripts/issue1482_*.py: best_pc_alignment, continuous_predictors, dense_predictor_reads, discrete_predictors, footprint_moments, jspace_r2, label_reads_densesae, lens_reads, pc_dashboard, pc_deviation_dashboard, predictor_battery, predictor_battery_fullwidth, rb7_reads, rb_as_sae_feature, result2_assembly, shapley_blocks, tail_depth_sweep

2 × scripts/issue1738_*.py: deviant_band, pc_sae_alignment_vs_r2

The corrective form already exists in-repo

scripts/issue1482_densesae_fullwidth.py:51-58 carries the fixed shape with its rationale comment:

# Derived from __file__, NOT task_workflow.repo_root(): that resolver REFUSES a ...
PROJECT_ROOT = Path(__file__).resolve().parent.parent
sys.path.insert(0, str(PROJECT_ROOT / "scripts"))

Note the parents depth differs between the scripts/ shape (.parent.parent) and the tests/ shape (.resolve().parents[1]) — do not copy the tests/ form blindly.

Why this was NOT absorbed into #2181

Landing the check with scripts/ in scope would have put 19 immediate offenders into the no-flags default bundle, which gates the Step 9c compare and the Step 10d pre-push merge for every task — an instant fleet-wide red. The remediation has to land before the scope widens. #2181 pinned the tests/-only scope with a green test case (case 13) precisely so this widening is a deliberate, reviewed step rather than an accident.

Acceptance criteria

1. All 19 drivers use the __file__-derived form; each still runs (import-level smoke at minimum).
2. check_no_repo_root_syspath_in_tests widened to scripts/ (rename to drop _in_tests), with the existing 15 test cases preserved and a scripts/-scope firing case added — the current case 13 pins scripts/ as OUT of scope and must be inverted deliberately, not deleted silently.
3. Scoped check green on the live tree; the no-flags bundle shows no new failures vs the plan-time baseline (the bundle is currently rc=0 / PASS / 0 FAIL / ~19-21 WARN, the WARN count varying by two transient root-guard hook lines that are environment- not tree-dependent).
4. src/ is clean today (0 hits) — widening to src/ as well is optional and should be justified separately if taken.

Also worth folding in (surfaced during #2181, unfixed)

• A select_step9c_tests.py argv dry-run in verify_plan.py. verify_plan.py c46 already dry-parses plan-embedded dispatch_issue.py commands; #2181's plan §6 shipped a --map-files invocation that ERRORs as written (the flag takes a newline-delimited path-LIST file, not positional source paths). A sibling dry-parse would have caught it at plan time.
• The pre-existing sibling ValueError gap at workflow_lint.py:11846, noted by #2181's code review. Investigated there and found NOT reachable on CPython 3.11.15 for the null-byte case (ast.parse raises SyntaxError, already caught) — carried here only so the note is not lost, not as a known defect.

Provenance

Deferred from #2181 (plan v3 §8) as a named follow-up. Filed WITHOUT auto-dispatch: this is a planned remediation, not a ripe bug fix, and no one has asked for it to run yet — leave at proposed for user/PM triage.