Inverter AC output power. Kept separate from active_power_kw because on a site with both an inverter and a revenue meter the two are different measurements of different things, and collapsing them makes inverter efficiency uncomputable.
Inverter AC output power.
inverter_output_kwEach of these is a real tag as the device or controller emits it, taken from the Forge mapping corpus. This is the table that does not exist anywhere else — the reason cross-vendor telemetry is hard is visible in a single column.
| OEM / profile | Raw tag as the device emits it | Confidence |
|---|---|---|
| Fronius | WDCWAC_WP_ACACPowerAC_Power+1 more spelling | 1.00 |
| SunSpec 101 | WAC_WP_ACPowerACPowerAC_Power+1 more spelling | 1.00 |
| SunSpec 103 | WAC_WP_ACPowerACPowerAC_Power+1 more spelling | 1.00 |
| Victron | Pv/Power/Dc/Pv/Power | 1.00 |
| SolarEdge (inverter) | I_AC_Power | 1.00 |
| Sungrow | Total DC power | 1.00 |
Every row above is an exact match — a deterministic pack entry or a confirmed corpus mapping — so all of them carry confidence 1.00. That is not a rounding: these are the two exact-match layers, and a tag that resolves here never reaches the embedding or LLM layers. Mappings resolved by the LLM layer do carry sub-1.0 confidence, but those are per-tenant and are not published here.
The canonical field is not just a name. It is a contract about units, sign convention and plausible range, enforced on the way in and re-checked at read time.
| SI / canonical unit | kW |
| Accepted input units | MW W hp kW |
| Converted on ingest | MW W hp → kW |
| Physics bounds | 0.0 to 100000.0 kW |
| Measurement type | instantaneous |
No mapping table to maintain, and no per-vendor branch in your code. The same request shape works for every OEM in the table above.
Read alongside this one.
spindle_speed_rpm spindle_load_pct program_name part_count feed_rate