Entities¶

Every price is the bare day-ahead exchange price in EUR/kWh, without grid fees, taxes or supplier margin, except the total price. The entities belong to one device per zone, Spot Forecast DE-LU.

Entity ids are fixed and English on every instance language, so the cards and other integrations can rely on them: a German instance also gets sensor.spot_forecast_de_lu_price, only the display names are translated.

Price¶

sensor.spot_forecast_de_lu_price

State: the price of the current slot (a known market price).
prices: every slot from today 00:00 to the end of the forecast, known slots first, forecast slots right after them, without a gap or overlap.
unit: EUR/kWh.
issued_at: when the shown forecast was made (ISO time).
anchor_day: the latest day whose prices were known when it was made; the forecast covers the six days after it.
model_version: the model that made it, e.g. de_lu-2026-10.
stale: true when the last run failed or is more than 7 hours old.
surcharge: the option grid fees and taxes in EUR/kWh, 0 when unset; the forecast card reads it for its "incl. fees" view.
vat: the option VAT on the market price in percent, for the same view. Unless set, the zone's rate: 19 (DE-LU), 21 (NL), 25 (DK1, DK2, NO2).
attribution: the data sources and their licences.

An entry of prices:

{"time": "2026-10-17T18:00:00+02:00", "price": 0.2134, "p90": 0.3812, "p10": 0.142, "source": "forecast", "h": 2}

Key	On	Meaning
`time`	every entry	Slot start, local time with offset. The slot ends where the next one starts.
`price`	every entry	Known price (`market`) or expected price, the median forecast (`forecast`).
`source`	every entry	`market` for a published price, `forecast` for a forecast hour.
`p90`	forecast	Peak risk: a price this high or higher is expected in about 1 hour in 10. Never below `price`.
`p10`	forecast	Low-price value: a price this low or lower is expected in about 1 hour in 10. Never above `price`.
`h`	forecast	Horizon in days after `anchor_day` (1 … 6).

Known slots keep their native resolution: 15 minutes since 1 October 2025, hourly before. Forecast slots are hourly. The attribute can hold around 300 entries, so it is not recorded (it would exceed the recorder's 16 KB limit); its history lives in the statistics instead.

A template for the lowest expected price among the forecast hours, as an example:

{% set prices = state_attr('sensor.spot_forecast_de_lu_price', 'prices') or [] %}
{% set ahead = prices | selectattr('source', 'eq', 'forecast') | map(attribute='price') | list %}
{{ ahead | min if ahead else none }}

Total price¶

sensor.spot_forecast_de_lu_total_price

Only while the option grid fees and taxes is set; setting it back to 0 removes the sensor.

State: the current exchange price with VAT, plus the surcharge: price × (1 + vat / 100) + surcharge.
prices: the same entries as on the price sensor, with price, p90 and p10 turned into total prices the same way. Smart Battery Pilot reads it like the price sensor.
surcharge, vat, unit, issued_at, anchor_day, model_version, stale, attribution: as on the price sensor.

Peak risk¶

binary_sensor.spot_forecast_de_lu_peak_risk

On when a price at or above the peak threshold is certain or likely in the next 48 hours: a known market price at or above it, or a forecast hour whose peak risk (P90) is at or above it. Counting both keeps the meaning stable across the day-ahead publication: when tomorrow's peak becomes a known price, the sensor stays on.

Attributes: slots (the flagged entries of prices), threshold (EUR/kWh) and certain (true when at least one flagged slot is a known price).

Negative price¶

binary_sensor.spot_forecast_de_lu_negative_price

On when a negative price is certain or likely in the next 48 hours: a known price below zero, or a forecast hour whose low-price value (P10) is below zero. Attributes as for peak risk.

Days¶

sensor.spot_forecast_de_lu_day_0 … sensor.spot_forecast_de_lu_day_6, disabled by default. Day 0 is today. State: the mean price of the day, weighted by slot length. Attributes: date, min, max, p90_max and p10_min (forecast slots only) and source: market when every slot of the day is a known price, forecast when every slot is a forecast, mixed when the day has both, and null with the other values when the day has no slots. They make automations such as "is Thursday cheaper than Wednesday?" easy.

Diagnostics¶

Entity	State	Attributes
`sensor.spot_forecast_de_lu_last_run`	Time of the last successful run
`sensor.spot_forecast_de_lu_model_version`	Active model version	`source` (`bundled` or `downloaded`), `train_cutoff`, `metrics` (the model's backtest report)
`sensor.spot_forecast_de_lu_forecast_error`	Mean absolute error of the expected price over the last 30 days, all horizons, EUR/kWh	`mae_h` (per horizon), `peak_mae` (hours above the threshold), `n`
`sensor.spot_forecast_de_lu_backfill_status`	`off`, `pending`, `running`, `done` or `failed`	`days`, `imported`, `recomputed`, `finalized_days`, `last_error`

The forecast error is computed from the integration's own stored forecasts, backfill included, against the prices that came. Compare it with the figures on the accuracy page.

Long-term statistics¶

Every forecast the integration makes is kept until its target day's prices are known. Once they are published, around 13:00 the day before, it writes that day's hourly external statistics (EUR/kWh); the forecast rows are not held back until the day is over:

Statistic id	Content
`spot_forecast:de_lu_price`	Actual day-ahead price, hourly mean
`spot_forecast:de_lu_forecast_h1` … `_h6`	Expected price, forecast with anchor = target day − h
`spot_forecast:de_lu_peak_risk_h1` … `_h6`	Calibrated P90 of the same forecast
`spot_forecast:de_lu_low_h1` … `_h6`	Calibrated P10 of the same forecast

If several runs share an anchor day, the last one wins: the forecast you saw last. Backfilled days are written the same way. A newer model only affects new runs; what was written stays.

The statistics appear in Developer tools → Statistics and can be used by any card that reads statistics, such as the look-back card or the built-in statistics graph card.