get_port_activity_report

Build a per-port runtime activity report from historical data.

Args: device_id: The AC Infinity device code (from discover_devices) days: Number of days to analyze. Default: 7. Must be 1–30.

Returns: JSON with window_start_local and window_end_local (the exact local time range analyzed, e.g. 'May 23, 10:35 AM CDT' to 'May 24, 10:35 AM CDT'), per-port on_hours (total hours ON over the full period), off_hours, transitions, avg_speed_when_running, uptime_pct, and peak_hour_local (device-local time string with peak date, e.g. '3:00 PM CDT (peak on May 23)', or null if the port never ran). Note: data_quality is an internal classification field stripped from the JSON output before serialization — it is NOT present in the response JSON. Its effects are visible only in human_summary: toggle hardware (heaters, lights, humidifiers — loadType 4 or 128 on standard devices, or pattern-detected on devType=18/22 where loadType is unreliable) produces a ▎-prefixed caveat line; devType=22 (Q0KT4 Genetics Lab) produces a device-level Note about missing power-draw data. devType=18 (UIS 69 Pro+) does NOT emit this Note — its active ports produce reliable runtime data in historical records even though portsLoad is always 0. ports_excluded_count is the number of ports removed by the ghost-port filter, capped at devPortCount when the device's physical port count is known (prevents over-counting on sub-8-port devices; unknown/zero devPortCount means no cap). Six rules apply: Rule A (constant 100%% uptime + zero load), Rule B (auto-named Port N with low average runtime or zero load), Rule C (named port with zero transitions + zero load + < 1 h/day average runtime), Rule D (non-toggle named port with speed history ≤ 1 and zero load — confirmed toggle hardware with transitions > 0 is exempt; see Quirk 22 in docs/API.md), Rule E (named port, non-toggle hardware, zero current load, sub-threshold runtime — stale configured speed from a port previously set to OFF), and Rule F (phantom clone detection — custom-named ports sharing identical activity signatures with low average on-time are excluded as legacy controller artifacts; fires only when port_loads data is available; proper-subset guard ensures at least one port is always retained). The human_summary field already includes a brief note about excluded ports when ports_excluded_count > 0. Do not repeat the exclusion count in prose response. The transitions count uses debouncing (_MIN_DWELL_READINGS=2): single-reading state changes at automation window edges are not counted — only transitions where the new state persists for ≥ 2 consecutive readings are recorded.

Ports whose timing data is unreliable appear only as ▎-prefixed caveat
lines in human_summary grouped by current state, e.g.
"▎ Currently ON: Heater (Port 2)." or
"▎ Currently OFF: Humidifier (Port 3)."
Do NOT quote on_hours or uptime_pct for these ports —
relay the caveat lines verbatim instead.

All ports listed under the main runtime sentences have reliable timing data
and should be presented normally. When a device-level Note about missing load
data appears in human_summary (devType=22 devices only), relay it once — do not
add further caveats.

Presentation guidance: - Always refer to ports as 'Name (Port N)', e.g., 'Exhaust Fan (Port 3)'. - When presenting on_hours to a grower, translate it from raw hours to natural language, e.g.: "The fan ran for 36.0 hours over the past 3 days (about 50% of the time)." Do NOT describe on_hours as hours per day. - window_start_local and window_end_local show the exact analysis window in the device's local timezone. Use these when explaining why a device shows activity from multiple calendar days (the window is a rolling 24h/N-day span, not a calendar-day boundary). - peak_hour_local is in device-local time with the peak date, e.g. '3:00 PM CDT (peak on May 23)'. - Ports with a ▎ caveat line: relay the caveat verbatim, no runtime numbers.