Evidence — unit rows (T1–T5, T14)

Rows T1–T5 of testing-plan.md, each run as its own command so a row's outcome is attributable. T14 (a poller with no heartbeat) has no suite of its own — it is a case inside T2, named in the run below.

Nothing captured here contains a credential, a hostname or personal data: these tests touch only temporary directories, and the poller holds no token.

T1 — heartbeat (cli/tests/test_poll_heartbeat.py)

$ uv run --project cli python -m pytest -q cli/tests/test_poll_heartbeat.py
........                                                                 [100%]
8 passed in 0.04s

Covers the round-trip, a started-but-not-cycled poller, startedAt staying put across cycles, error counting and interruption, atomic replacement leaving no .tmp behind, an absent/corrupt/wrong-shaped file reading as None, and — the one that matters for ingress — an unwritable path warning once and never raising.

T2 — poll status (cli/tests/test_poll_status.py)

$ uv run --project cli python -m pytest -q cli/tests/test_poll_status.py
..........                                                               [100%]
10 passed in 0.52s

Includes the two abuse cases of row T7 and the migration case of row T14:

Test	What it pins
test_a_forged_heartbeat_cannot_make_a_dead_poller_look_alive	liveness is the lock; a heartbeat with no poller behind it reads not running, exit 1
test_a_stale_pidfile_is_reported_and_left_alone	a stale pidfile is named, and not deleted by a read-only command
test_heartbeat_absent_still_reports_liveness	a poller from before the heartbeat existed still reports running and its pid (R4.8 / T14)

T3 — argument wiring (cli/tests/test_poll_command.py)

$ uv run --project cli python -m pytest -q cli/tests/test_poll_command.py
.................                                                        [100%]
17 passed in 0.76s

Ten pre-existing tests plus seven for this work item: --daemon/--foreground last-one-wins, the three paths defaulting under state.root, --daemon --once exiting 2, the pre-fork refusals (lock held, logfile unopenable) both proving no fork happened, a stale pidfile removed by the next start, the heartbeat written by a completed cycle, and daemon_entry forcing foreground.

T4 — state classification (cli/tests/test_state_portability.py)

$ uv run --project cli python -m pytest -q cli/tests/test_state_portability.py
.......                                                                  [100%]
7 passed in 0.02s

The suite is unchanged; it now covers three more paths by construction. Before the documentation and .gitignore were updated it failed exactly as designed — the run that proved the gate works:

FAILED test_every_generated_path_is_documented - AssertionError: poller pidfile
  (.the-loop/poll.pid) is missing from the classification table in docs/cli/state.md
FAILED test_block_ignores_exactly_the_local_paths - AssertionError: poller heartbeat
  (.the-loop/poll-status.json) is declared local but the documented .gitignore block
  tracks it

T5 — control plane (cli/tests/test_core_daemons.py)

$ uv run --project cli python -m pytest -q cli/tests/test_core_daemons.py
.......                                                                  [100%]
7 passed in 0.06s

daemon_status("poller") carries startedAt/lastCycleAt from the heartbeat and gh-webhook carries them empty; control_daemon("poller", "start") opens the logfile before spawning and passes the same fd as stdout and stderr — the DEVNULL that used to discard a control-plane start's output is gone.