Decide what tests/runtime_smoke.py is: pytest never collects it, it writes tmp_smoke_test/ into the cwd, and testing.md misdescribes it #912

Deferred work · F-PR29-A4-05 · severity Low · cost ~1 h · recorded in docs/roadmap/risks-and-decisions.md:83

Summary

tests/runtime_smoke.py is the only script that runs a real one-step training (SmolLM2-135M), but its name keeps pytest from collecting it, so it never runs. It writes into ./tmp_smoke_test with the cleanup commented out, carries Turkish comments, and testing.md presents it as a dry-run fixture. Options recorded at PR #29: move it to tools/manual_smoke.py as a documented release gate, or rename it with mocked model loading and tmp_path; folding it into the nightly smoke #106 asks for is a third.

Condition or budget

None recorded (the row still names v0.6.x).

State today (re-checked 2026-10-09)

The file is unchanged since 2026-04-26; pyproject.toml sets only testpaths, so it is not collected.

Related

#106 (CR-A16-022) asks for an unmocked nightly smoke per trainer_type; it names this file but not its disposition.

Acceptance criteria

• The file is deleted, folded into the nightly smoke, or moved to tools/ as a documented release-gate check — no uncollected script is left under tests/.
• Nothing writes into the working directory, comments are English, and testing.md describes what exists.
• The row is removed from the deferral table and the CHANGELOG names its ID (the table's removal contract).

_{Recorded in docs/roadmap/risks-and-decisions.md (read at f94595f) and re-checked against the code on 2026-10-09. The row leaves that table only when the fix is on main, the CHANGELOG names its ID and any promised guard exists.}