PlausibilityValidator

Checks that values lie in plausible ranges and that they don’t stay frozen for too long.

Basic Usage

from epftoolbox2.data.validators import PlausibilityValidator

# Defaults: prices within the SDAC limits, load at least 1 MW, price/load frozen for over 24 rows
result = PlausibilityValidator().validate(df)

result = PlausibilityValidator(
    ranges={"price*": [-600, 4000], "temperature*": [-40, 50], "generation_*": [0, None]},
    max_constant_run=24,
    constant_columns=["price", "load_actual"],
).validate(df)

Parameters

Parameter	Type	Default	Description
ranges	dict or None	None	{column or glob pattern: [min, max]}, bounds inclusive, None for no bound. None uses the defaults below. {} turns the range check off.
max_constant_run	int or None	24	Longest run of identical consecutive values allowed, in rows. None turns the check off.
constant_columns	List[str] or None	None	Columns or glob patterns checked for constant runs. None uses price* and load_actual*. ImputeTransformer flag columns (*_imputed) are always skipped.

A column uses the first pattern in ranges it matches, so put specific patterns before general ones. Only numeric columns are checked. A pattern you pass that matches no column gives a warning.

Defaults

Pattern	Range	Why
price*	-600 to 4000 EUR/MWh	SDAC harmonised minimum and maximum clearing prices. The maximum has been 4000 since 2022. The minimum was -500 until trading day 2026-05-28 and -600 since.
load*	at least 1 MW	Load can’t be zero or negative. Some TSOs report zeros during outages.

The SDAC limits move automatically under ACER’s HMMCP methodology when prices come close to them. Pass ranges explicitly if they change, for non-SDAC markets, or for prices in another unit or currency.

Constant runs are checked only in price* and load_actual* by default: calendar dummies, daily values upsampled to hours and night-time solar output are constant by design.

Output

• Errors: values outside their range (count, first timestamp, min and max).
• Warnings: runs of identical values longer than max_constant_run. They are warnings because a price can genuinely repeat for hours, but a long run usually means a filled outage or a frozen feed.
• info: out_of_range (per column), longest_constant_run (per column), checked_columns.
• stats: one row per checked column: rule, low, high, min, max, out_of_range, longest_constant_run, constant_value, constant_start.

print(result.warnings[0])
# Column 'price' repeats 92.54 for 30 consecutive rows from 2024-01-03 01:00:00+01:00
# to 2024-01-04 06:00:00+01:00 (more than 24)

When to Use

• On raw source data, to catch parsing errors (wrong unit, a mixed-up series) and outages filled with a constant.
• After resampling, where a flat run is often what interpolation or forward fill left behind.