Summary

nova.DerTelemetry unmarshals a driver's emit blob directly, so a key it does
not recognise is dropped without an error and the value never leaves the
gateway.

Catalog drivers emit import_wh, export_wh, charge_wh, discharge_wh,
lifetime_wh and hz. DerTelemetry reads total_import_wh,
total_export_wh, total_charge_wh, total_discharge_wh,
total_generation_wh and freq_hz.

None of those six has ever reached Nova from a catalog driver. Energy
counters and grid frequency, from every driver in srcfl/device-drivers.

Our own zap.lua works around it per field — reading.import_wh = reading.total_import_wh — which is why the gap went unnoticed: the drivers
written here happened to do the right thing.

What changed

normalizeTelemetryKeys rewrites known aliases onto the names the payload
reads, at the emit boundary, so every driver is fixed at once.

The canonical @srcful/data-models spellings map onto the same internal names,
so srcfl/device-drivers can convert its catalog to W, Hz, L1_V,
SoC_nom_fract and total_import_Wh without a further host change.

Three rules keep it safe:

• an alias never overwrites a key the driver set under the target name;
• the driver's original keys are kept, because the UI and API surface a reading
  verbatim and removing one would change what an operator sees;
• a parse failure returns the input untouched — the envelope is already
  validated by this point, and normalisation must never lose a reading.

Tests

go/internal/drivers/telemetry_keys_test.go:

• catalog spellings (hz, import_wh, export_wh) land on the Nova names;
• battery and PV counters (charge_wh, discharge_wh, lifetime_wh) likewise;
• canonical spellings (W, Hz, L1_V, total_import_Wh) land on the same
  names;
• a driver that already sets the target name keeps its own value;
• a payload carrying both names passes through byte-identical, so FTW's own
  drivers are unaffected;
• malformed input is returned unchanged.

go test ./internal/drivers/ ./internal/nova/ ./internal/telemetry/ passes.

Follow-up

rated_w → rated_power_w and the battery v/a → dc_v/dc_a mappings
are deliberately left out. Both need a look at what the drivers actually mean
by those fields before mapping them, and neither is an energy counter that
silently reads as zero.

🤖 Generated with Claude Code