Schema engine for energy market data
The forecasting database for energy markets.
Pigface keeps every version of every market table and serves one history, correct as of any date. Generators, networks, insurers and startups train and forecast on all of it. Millisecond queries, nothing dropped.
- tables in the model
- 571
- archive start
- 2009
- rows dropped
- 0
Latest data model: v5.7, May 2026.
Keep. Read. Query.
-
01 · Keep
Every file the market ever published.
Kept exactly as it arrived, with where it came from recorded. Nothing is rewritten, nothing is thrown away.
-
02 · Read
Every version of every table.
Old shapes and new shapes read the same way. A value that does not fit is kept, not dropped.
-
03 · Query
One history, any date.
Ask for last March and get last March as it was published. Parquet underneath, so DuckDB, Polars and your notebooks already work.
-- one table, every version, one query
SELECT interval_start, duid, scadavalue
FROM read_parquet('lake/nem/dispatch_unit_scada/**/*.parquet',
hive_partitioning = true)
WHERE interval_start BETWEEN '2024-01-01' AND '2024-02-01'
AND duid = 'BAYSW1';model training · forecasting
Train on the whole market.
The same history, served as a dense forecasting grid, a point-in-time feature store with no future leakage, or a temporal graph of units, regions, interconnectors and constraints.
Talk to us.
hello@pigface.io
No form, no tracking. Email, and a person answers.