Health

Every interval, output is divided by the sunlight the Tempest measured and the array's kilowatts. Before the Tempest went up in October 2024, Open-Meteo's hourly satellite estimate stands in. A day's score is that ratio over its sunny intervals (300 W/m² or more) against the median of the 30 days before it on the same sunlight source, so season, heat and the sensor's flat mounting cancel out.

A day is an incident when it scores under 75%, or when output stays under 60% of normal for an hour or more in strong sun (500 W/m² or more), which catches an inverter that trips for part of a day. Passing cloud the sensor sees and the panels don't gives runs of 15–30 minutes. Days with under two hours of sun aren't judged.

Since October 2026 both gateways are also read every minute. The Powerwall gateway meters the whole array; the expansion inverter's own gateway meters it alone, so the two original inverters can be told apart from the expansion. A sudden fall on one side while the other holds steady is counted as an inverter drop.

A bill is flagged when house use, the bill amount, solar or energy from the grid is at least 2 standard deviations from the average of the previous 12 bills (at least 3 needed), or when the bill's energy used, energy delivered or billed kWh differ from the battery system's import, export or net import over the same billing period by 10% and 25 kWh or more. The battery side is the difference of its lifetime meter counters across the period, so a missing day doesn't skew it. When billed kWh agrees but used or delivered don't, the bill figures are the likely problem. Saving a reason moves the bill to Reviewed; it stays there even if the check keeps firing.

Data health: a skipped day has no daily row, or a row with no totals, up to the last day whose row should exist; readings over 30 minutes apart are 15-minute rows closing a gap, drawn as gaps rather than averages; the energy balance is solar + import − export − house, flagged when it is more than 10% of house use (a battery loses a few percent, so measured bills sit near +4%); a rate jump is a bill whose rate moved more than 10% from the previous bill's (assumed rates are bills with no billed kWh); estimated or partly metered days are the ones hatched on the charts. Skipped days also raise a Slack alert.

Suggestions: capacity factor (solar_kwh ÷ (system_size_dc × hours)) under 13% for three bills; the last 12 bills' solar under 90% of the PVWatts projection for the same billing periods, at the array size of the time; average house use of the last three bills 15% above the same bills a year earlier, when all six had solar for the whole bill.