[frahlg-agents]

Reported by Anders (Growatt PV + Pixii meter/battery, v0.154.1): after sunset Growatt goes offline and /api/status shows pv_w:null, load_w:null, history_reason:"offline:Growatt:pv" although Pixii still reports the site meter.

Root cause — measurementLocked (go/internal/telemetry/forecast.go) fails the whole balance (Valid=false) whenever any expected flow's driver is !TelemetryLive() (reason = "offline:"). qualifyStatusHistory then nulls load_w (MissingLoad) and pv_w (MissingPV).

Fix (live path only, alignSources == false): a PV source that is offline or stale and whose last reading was within ±50 W (PVDarkThresholdW) is counted as 0 W. The source is marked AssumedZero, the reading gets PVAssumedZero; Known[DerPV] and PVValid stay false so history and PV forecast training do not record an assumed value as measured. /api/status shows pv_w: 0 and the load from meter + battery.

Unchanged: PV that drops out while producing (last reading > 50 W) still fails as offline:…:pv; a missing site meter still nulls load; a PV driver with no reading ever (missing:) still fails; forecast/history alignment paths unchanged.

Tests: new cases in telemetry/live_measurement_test.go and api/api_history_quality_test.go. go test green for telemetry, api, ha, state, cmd/ftw. Locally internal/drivers fails only because the bundled driver catalog is not checked out in my worktree (unrelated).

Open questions: should the night check use internal/sunpos (sun below horizon) instead of / in addition to the last-reading heuristic? Is 50 W the right threshold?

[frahlg-agents]

Closing in favour of #1768, which restores live Home use from the site meter with quality tracked separately (no 50 W heuristic).