After restoring NORMAL mode from persistent storage (line 264), turnOnComponents() is called which turns on face switches (0-5). However, the code doesn't log any event to indicate this state restoration occurred, unlike SAFE_MODE (line 252-253) and PAYLOAD_MODE (line 260-261) which log their restoration. This inconsistency means operators won't see any event confirming NORMAL mode restoration after a reboot, making it harder to track system state changes. Consider adding a log event similar to the other modes.

The error check [ -e "$$TARGET" ] uses the -e test which checks if the file exists, but this will succeed for directories too. If someone passes TEST=int, it will find the directory FprimeZephyrReference/test/int and incorrectly proceed to run pytest on the directory instead of failing with "file not found". Change to [ -f "$$TARGET" ] to specifically check for a regular file, or adjust the logic to handle directories explicitly.

The comment "// Called from run_handler at 1Hz when in hibernation wake window" is helpful, but it would be beneficial to document what happens when the wake window expires. Specifically, that enterDormantSleep() is called again, which will increment counters and potentially cause another reboot cycle.

Consider expanding the comment to: "// Called from run_handler at 1Hz when in hibernation wake window. When counter reaches wakeDurationSec, automatically re-enters dormant sleep for the next hibernation cycle."

The docstring for pytest_addoption says "Add custom command-line options for integration tests" but doesn't explain what the --run-reboot flag does or when it should be used.

Consider expanding the docstring to:

"""Add custom command-line options for integration tests.

--run-reboot: Enable tests that trigger actual system reboots. These tests
are skipped by default because they require physical hardware and cause
the system to restart. Only use on real hardware for hibernation testing.
"""

In the loadState() function, when the read fails or has a size mismatch (line 277), the code initializes to default values and calls turnOnComponents(). However, this could be problematic if the system was actually in HIBERNATION_MODE before the read failure.

If a read error occurs during hibernation wake window restoration, the system will:

1. Default to NORMAL mode (line 280)
2. Turn on all face switches (line 287)
3. Not properly initialize hibernation state variables (m_inHibernationWakeWindow remains false from constructor)

This means that if the persistent state file becomes corrupted during a hibernation cycle, the system will incorrectly restore to NORMAL mode with all switches on, potentially violating power constraints that hibernation mode was trying to maintain.

Recommendation: Consider logging a critical warning when state restoration fails, and possibly defaulting to SAFE_MODE instead of NORMAL to maintain a conservative power profile.

The M33_SCR_SLEEPDEEP_BITS constant is used without being defined or included. This is likely a Cortex-M33 specific constant that should come from CMSIS headers. Ensure this constant is properly defined, or use the standard SCB_SCR_SLEEPDEEP_Msk from CMSIS core_cm33.h instead.

The comment states "Hibernation counters (cycleCount, totalSeconds) were already incremented in enterDormantSleep() BEFORE the dormant sleep and saved to persistent storage. We're restoring those pre-incremented values." However, this is slightly misleading. The counters are incremented in enterDormantSleep(), but only if the system successfully enters dormant mode and reboots. If dormant entry fails, the counters are rolled back. Consider clarifying this comment to mention that these are the "successfully saved" values after dormant entry, not necessarily pre-incremented values that might have been rolled back.

The code calls go_dormant() and then expects execution to continue at line 202 with aon_timer_disable_alarm(). However, according to the implementation at line 211, after waking from dormant, the code immediately calls sys_reboot(SYS_REBOOT_WARM) which never returns. This means lines 202-205 (aon_timer_disable_alarm() and sleep_power_up()) are unreachable code and will never execute.

Either remove these unreachable lines or restructure the code so that the alarm disable and clock restoration happen before the reboot decision is made.