workflow-fix: step9c --map-files misses tests importing/source-inspecting a changed src module

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a prose follow-up raised on task #1112 (emitting agent: experiment-implementer, rankem round).

Goal

Make select_step9c_tests.py --map-files map tests that statically import a changed src/ module (and consider symbol-level source-inspection dependencies), or fail loud when a src/ path yields zero mapped tests.

Workflow gap

• Bug observed: --map-files src/explore_persona_space/train/sft.py returns an empty test mapping while tests/test_artifacts_recipe.py imports train_lora from that module (line 47) AND inspect.getsource-inspects it (line 553); a use_rslora change to train_lora (commit d7908a3837) broke test_artifacts_recipe.py::test_rslora_engine_pin and the selector never surfaced it — the implementer caught it only by running sibling tests directly.
• Why it is a workflow gap: the Step-9c mapped-scan gate and the inline payload lint gate both rely on this selector; a round can push src/ payload that breaks the fleet-wide test gate without either gate firing. The miss is not only the exotic source-inspection case — a plain static import edge was not mapped.
• Confidence (emitter): low (raised to medium by filer verification — the miss reproduces).
• verified-at-filing: uv run python scripts/select_step9c_tests.py --map-files src/explore_persona_space/train/sft.py → 0 mapped tests (header only); grep -n 'getsource\|train_lora' tests/test_artifacts_recipe.py → import at :47, inspect.getsource(train_lora) at :553; grep -rln 'inspect.getsource' tests/ → 10+ test files use source inspection (breadth of the symbol-level blind spot); git log --oneline --since='7 days ago' -- scripts/select_step9c_tests.py → 5 commits, none adding src-module import mapping (subjects: staged-index verification #1346, guard-surface rounds #1338, digest crash-tails #1316, ruff pin #1307, rules-pin discovery #1270) (2026-07-20).

Proposed change (candidate diff sketch — refine in planning)

(none — synthesized from prose follow-up; the emitter's sketch in words: extend the mapper's dependency edges from filename/import-graph heuristics to (a) tests importing the changed module path and (b) tests source-inspecting a changed symbol; alternatively a fail-loud "0 tests mapped for a src/ change" warning so the gap is visible at gate time.)

Scope / surfaces

• Primary target: scripts/select_step9c_tests.py
• Grep the workflow surface for the pattern before editing (grep -rln 'map-files\|map_files' .claude/ CLAUDE.md scripts/) 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.
• This session runs under EPM_WORKFLOW_FIX_SESSION=1 and carries a workflow_fix_target: Provenance line — it MUST NOT auto-route any of its own subagents' workflow-fix candidates (recursion guard).

Provenance

• workflow_fix_target: scripts/select_step9c_tests.py
• fingerprint: 219c0bffd790

Verbatim surfaced prose (experiment-implementer report, task #1112 rankem round, 2026-07-20):

scripts/select_step9c_tests.py --map-files src/.../train/sft.py returned EMPTY, yet my change to train_lora broke test_artifacts_recipe.py::test_rslora_engine_pin, which inspect.getsource(train_lora). The mapped-test oracle misses tests that source-inspect/import the changed SYMBOL (not just import-graph/naming). Backstop (running sibling tests directly) caught it — low confidence, but worth a look for the step9c selector.