workflow-fix: step-9c selector import-map arm (stem-map misses import-relationship tests)

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a workflow-fix candidate raised on task #1286 (emitting agent: #1286's Phase-2 statistics critic, surfaced prose — "Selector-shape observation").

Goal

Extend scripts/select_step9c_tests.py so the touched-scope selection also picks test files whose module-top imports name a touched scripts/ module (import-graph mapping, or an explicit module→test registry), so touched-module test coverage no longer depends on test-file NAMING.

Workflow gap

• Bug observed: the touched-scope stem-map selector misses tests that exercise a touched module under an unrelated filename. Concrete instance (#1286): tests/test_issue810_uh_pack_validation.py imports issue810_common, issue810_fit_readout, and issue810_bootstrap_deltaskill, yet when exactly those scripts are the touched set, select_tests_with_reasons() does NOT select it — the stem globs (tests/test_{stem}.py exact + tests/test_*{stem}*.py broad) match nothing because "uh_pack_validation" contains no touched stem. A code change to those modules can therefore pass the step-9c gate without its most relevant test ever running (4 of the 5 touched scripts land only in the untested_touched WARN list). #1286's plan had to add a manual "append the target test explicitly to the gate invocation" binding as a workaround.
• Why it is a workflow gap: select_step9c_tests.py is the single source of the step-9c touched-scope selection (workflow surface); its stem-map maps by FILENAME similarity only, with no import-relationship arm, so the gate's coverage silently varies with test naming conventions.
• Confidence (emitter): medium
• verified-at-filing: grep -n "stem" scripts/select_step9c_tests.py → stem-map documented at lines 24-34 (exact + broad *{stem}* glob arms, no import arm) (2026-07-13); the #1286 statistics critic re-ran select_tests_with_reasons() on the 6-file touched set the same day and confirmed the target test absent from the selection.

Proposed change (candidate diff sketch — refine in planning)

+ In select_tests_with_reasons(): after the stem-map pass, add an
+ import-map pass — for each tests/test_*.py, parse module-top
+ `import X` / `from X import ...` statements (ast, cheap) and select
+ the test when X's resolved file is in the touched set; reason string
+ `import-map:<touched file>`.
+ Keep the stem map + GLOB_SCAN_TESTS arms unchanged; extend
+ recommended_timeout_s() sizing to count import-map selections.
+ Pin behavior in tests/test_select_step9c_tests.py (the #1286 shape:
+ touched issue810_common.py must select test_issue810_uh_pack_validation.py).

Scope / surfaces

• Primary target: scripts/select_step9c_tests.py
• Grep the workflow surface for the pattern before editing (grep -rln 'stem' scripts/select_step9c_tests.py tests/test_select_step9c_tests.py) and update every hit; list them in the plan.

Constraints / invariants

• Workflow-surface only — never experiment code, configs/, or tasks/.
• scripts/workflow_lint.py --check-asks passes; ruff on touched files passes; if workflow.yaml or CLAUDE.md change, they stay consistent with the rule file.
• Selection may only GROW (safe-by-direction: over-selection costs minutes, under-selection ships regressions); the import-map arm must never remove a stem-map/GLOB_SCAN selection.
• Keep select_step9c_tests.py --map-files (the #1147 lint-gate leg contract) byte-compatible.

Provenance

• workflow_fix_target: scripts/select_step9c_tests.py
• fingerprint: c4e9af27d15f

Surfaced prose (verbatim, from #1286's Phase-2 statistics-critic report): "Selector-shape observation (for the orchestrator, no candidate block — AUTO_REVIEW_DISABLED): select_step9c_tests.py's stem-map misses tests that exercise a touched module under an unrelated filename — exactly the fleet-breaking test here."