TimerMeet regression testing

Overview

TimerMeet's core value proposition is alarm reliability (see SDD.md's rationale for the PHP->Python rewrite). Regression testing here is not generic "does it run" -- it's specifically about not silently weakening an alert path. Pair with the timermeet-qa-validator agent, which owns the PASS/FAIL/BLOCKED reporting format.

What to validate, and how

Area	How to check
App startup	python timermeet.py, confirm no exception lands in data/timermeet.log, window appears within a few seconds (no freeze -- see design-notes.md's startup-freeze history).
Windows/navigation	List/gadget/calendar/week views switch via set_active_view/set_gadget_mode without a second Tk()/Toplevel appearing; each view's own nav (month/week prev-next, today button) still moves the displayed range.
Buttons	Grep the button's command= target in main_window.py/app.py to confirm it's still wired to a real handler, not orphaned by a rename.
Alarms	python -m unittest tests.test_alarm_queue -v (covers the FIFO queue/hand-off logic with a real tk.Tk() root); manually trigger one (shortest reminder, setNowButton) to see the overlay, its flash, and both buttons.
Sound	Confirm audio.AlarmPlayer.play still has its winsound.Beep fallback intact if you touched audio.py -- a failing MP3 must never mean silence.
Timers	Confirm every root.after(...) job your change touches is cancelled via after_cancel on its teardown path (alarm_ui.py::AlarmController.dismiss is the reference example -- it cancels _flash_job/_title_blink_job/_relift_job/_countdown_job in one loop).
Teams	Any webbrowser.open() call site goes through security.is_http_url() first (grep to confirm no new bare call site was added).
Calendar	python -m unittest tests.test_calendar_day_click tests.test_context_menu tests.test_week_view tests.test_week_selection tests.test_week_column_mode tests.test_header_layout -v if main_window.py's calendar/week rendering changed.
Notifications	notifications.py's OS toast is best-effort by design -- confirm it degrading/failing doesn't affect the sound/overlay channels (they must never depend on it).
Settings/persistence	python -m unittest tests.test_storage tests.test_merge tests.test_companies tests.test_retention -v if storage.py/retention.py changed -- these cover merge-on-save and the two cleanup functions, both real historical bug sources.
Clean shutdown	Close the app while no alarm is active, then again with one active (confirm dismiss(run_callback=False, advance_queue=False)'s shutdown path in alarm_ui.py doesn't try to build a new Toplevel against a dying root).

Full regression pass

python -m py_compile timermeet_app/*.py timermeet.py build_exe.py
python -m unittest discover -s tests -v
python timermeet.py   # manual smoke test; close after confirming no log exception

Never test against live data

Before running any script that calls handle_save/handle_delete/storage.save_meetings/etc. against real data, copy data/meetings.json first. A real meeting has been lost this way once already during testing -- see [[feedback_never_test_against_live_data]].

When something fails

Report the specific failing test/behavior and hand off to the timermeet-debugger agent for root-causing rather than guessing at a fix -- see that agent's diagnose-before-modifying discipline.