EIA Short-Term Energy Outlook (STEO)

It is the baseline everyone else argues with - the U.S. Energy Information Administration's standing forecast of energy supply, demand and prices, released around the second week of each month, delivered through Datadory as structured rows instead of a workbook download. Each edition pairs roughly two historical years with monthly estimates running through the end of the calendar year after next: about sixteen to seventeen months of forward visibility at any point in time. The August 2026 edition spans 2024 through 2027, forecast-complete as of August 6, with the next edition due September 9.

The span is wider than the oil patch alone. Global oil market assumptions sit beside the Brent and WTI price paths; world and non-OPEC production and consumption roll into US petroleum and hydrocarbon gas liquids balances, including refinery input and utilization; regional motor gasoline prices and inventories carry PADD-level detail; biofuels, natural gas, electricity, coal, macroeconomic indicators and drilling productivity round out the frame. Because every number comes out of one model run, the crude production, refining, consumption and inventory projections reconcile with each other and with the macro and weather inputs behind them - a coherence no pile of independently reported series can offer.

Twenty-eight tables, roughly 140 to 175 rows apiece, each row keyed by a series mnemonic - BREPUUS for Brent spot, DSWHUUS_$ for wholesale diesel - and a dedicated comparison isolates what moved between editions, so a revision arrives legible as data rather than rumor. An archive of past editions reaches back to 1983. Set inside the wider catalog - 1,744 datasets averaging a 7.81 quality score - the integrated oil & gas slice holds 23 datasets averaging 7.17, and this record scores 9/10: verified field definitions, complete forecast-horizon coverage and the only forward-looking window in the cluster. Get a sample of this dataset cut to the series you actually model.

Three cuts of one edition, exactly as the dictionary shapes them:

# one series, one period, one forecast vintage
table            : Table 2. Energy Prices
series_id        : BREPUUS
row_label        : Brent Spot Average
period           : annual columns, 2024 through 2027
value            : 81   69   87   69      # dollars per barrel
forecast_date    : 2026-08-06             # where history ends and projection begins

# same shape, product rack instead of marker crude
series_id        : DSWHUUS_$
row_label        : Wholesale Diesel Price (dollars per gallon)
value_2026_proj  : 3.37
value_2027_proj  : 2.62

# the revision row a desk actually reads
table            : Notable Forecast Changes (August 2026 vs July 2026)
row_label        : Wholesale gasoline price (dollars per gallon)
current_2026     : 2.91   previous_2026 : 2.75   change : +5.9%

Read the anatomy rather than the magnitudes. The first block shows four annual columns on one row - two of them history, two of them projection - so observed and forecast share a single shape and never need reconciling downstream, while the forecast date rides alongside marking exactly where the handoff happens. The second repeats the identical five-field shape for wholesale diesel, which is why a Brent path and a PADD gasoline inventory load into one table without reshaping. The third is the row a marketing desk actually reads: the 2026 wholesale gasoline view moved from $2.75 to $2.91 a gallon between the July and August editions, a 5.9 percent upward revision - interesting not as a footnote but as a series, once several editions are stacked.

That stacking is what a Datadory delivery makes cheap: keep the edition stamp on every pull and the revision panel builds as a diff query rather than a re-read of twenty-eight sheets. The rows above are illustrative of the delivered shape; the complete series set for your named mnemonics arrives with the sample.

Five working fields carry every series, all verified against source documentation during the August 2026 research pass. Series ID is the mnemonic naming the line - BREPUUS for Brent spot, WTIPUUS for WTI, RACPUUS for refiner acquisition cost, MGWHUUS_$ for wholesale gasoline, DSWHUUS_$ for wholesale diesel - stable enough across editions that a delivered history is one column rather than a stack of files. Row Label carries the human-readable description printed beside it, so a mnemonic and its meaning never travel apart.

Then the three columns that decide whether your model tells the truth. Period stamps each observation as an annual or monthly cell, totals for the history years and monthly values inside the projection window - filter on period type before mixing tenses, because a projected cell sitting unnoticed in a historical panel is the classic misread here. Value holds the figure in whatever unit the table header declares: dollars per barrel, dollars per gallon, million barrels per day, percent. Unit context stays attached to the row instead of hiding in a legend, which matters in a record where prices, volumes and percentages share one schema. Forecast Date records when the edition's forecast was completed, printed on every sheet - the vintage mark that lets cross-edition joins resolve cleanly.

Beyond the five sit the companion views - chart-gallery figures, cross-edition revision panels, unit metadata per series, Data Browser custom views - folded under additional fields on request in the table below rather than approximated here.

Geography - world, OPEC and non-OPEC aggregates alongside United States national series, with PADD-level regional detail across the five Petroleum Administration for Defense Districts for prices, inventories and electricity, so a Gulf Coast gasoline picture reads separately from a West Coast one while loading into the same field structure. International detail thins out fast beyond those aggregates: country-by-country resolution lives in a different record entirely.

Time frame - roughly two historical years plus monthly projections through the end of the following calendar year in every edition; the August 2026 tables cover 2024 through 2027. Past editions are archived as separate vintages reaching back to 1983, which makes this one of the longest continuous official forecast records in energy - though older months arrive as distinct snapshots rather than one seamless series, so long-run revision studies need the archive pulled deliberately.

Granularity - monthly and annual periods at national and regional cuts, one row per series per period, where a series is a fuel-times-flow-times-geography mnemonic inside the 28 numbered tables. Nothing is pre-blended: a seasonal gasoline view, a refining-margin proxy or a coal-vs-gas switch story all come off the same grid by filtering.

Measured against the catalog - 1,744 datasets averaging 7.81, the integrated oil & gas slice holding 23 averaging 7.17 - this record's 9/10 reflects verified definitions and a complete forecast horizon, docked mainly because its international depth stops at the aggregates. Where it ranks in the slice: best integrated oil & gas datasets.

API, files, or your warehouse. Daily, weekly, or hourly.

Pick the series you need - the Brent and WTI paths, refinery input and utilization, the PADD price sets, the inventory lines - pick the cadence, pick the landing zone. Every edition arrives parsed into the same five-field row shape whether it lands in Snowflake, BigQuery or Redshift or as flat files sized for an overnight load, so the August-over-July revision is a diff query rather than a re-read of twenty-eight sheets.

Normalization happens before anything reaches you. Workbook layout arrives as typed columns keyed by series mnemonic rather than report formatting, unit declarations ride with the values they describe, the projected-versus-historical split stays queryable off the period type, and every delivery includes the field dictionary above unchanged plus validation rows keyed to your named series. Name the series, regions and periods you model when you request the sample and it comes back scoped to them - get a sample of this dataset first and let the ongoing arrangement follow it.

Ranked by how directly a single row settles their day job:

1. Quant and fundamental investors. The consensus price and balance path under every energy thesis - history and projection in one row, ready for spread work and realized-versus-forecast scoring.
2. Data scientists and ML engineers. A five-field grid repeated across 28 tables with a forecast-completion stamp on every record, joining cleanly to realized-price panels on dates without provenance archaeology.
3. Market researchers and consultants. Client-ready outlook numbers from a named federal statistical agency, refreshed on a known monthly clock.
4. Journalists and analysts. When a story says output climbs next year, the projection row with its forecast date is the receipt.
5. Strategy and planning teams. Demand sizing and scenario baselines in stated units, so a growth claim commits to numbers before it commits to a deck.
6. Product and intelligence teams. Coherent fundamentals behind dashboards - one model run means the production, refining and consumption lines reconcile instead of arguing.

For contrast within the slice: EIA International Energy Statistics scores 9/10 answering what happened across 218 countries over decades, and the Baker Hughes Rig Count answers how many rigs turned this week. Neither claims anything about next year - this forecast owns that question.

Investors & Quant Researchers get the tightest fit - a consensus path with the vintage stamp their realized-versus-forecast panels require. Data Scientists & ML Engineers get a uniform grid that drops into a pipeline as-is, forecast state included as typed columns rather than footnotes. Market Researchers & Consultants get citable projections from a named public agency. Journalists, Academics & Students get receipts, plus an archive back to 1983 that turns today's consensus into a track record. Developers & Data-Product Builders get a schema stable enough that a new edition is new rows, never a migration. The constant is the shape: because history and projection share one five-field row, extending a panel across the forecast boundary is arithmetic rather than reconstruction. Persona-by-persona detail lives on the industry pages: investors and quants, data scientists, market researchers, competitive intelligence and product teams, developers and builders.

Inside this slice the datasets answer different questions, and the honest framing is windshield versus rearview mirror. EIA International Energy Statistics (9/10) answers how much was produced, where, going back decades across 218 countries and territories. JODI Oil & Gas World Database (8/10) answers what do the countries themselves say they produced, traded, refined and stored, month by month. Norwegian Offshore Directorate FactPages (10/10) answers what exists on the Norwegian shelf. This record answers the question none of them touch: what does the reference forecast say happens next - the only forward-looking window in the cluster, rolled forward every month out of one internally consistent model run.

Its edge deserves stating plainly: prices as well as volumes, projected coherently across fuels, with revisions documented between editions rather than silently overwritten. Pair it rather than substitute it - actuals from the statistics and regulator sources, the forward view from here. The head-to-head against its own sibling is laid out in eia international energy statistics vs eia short-term energy outlook.

Four things, stated up front.

First, tense discipline. Every value is either an observation or a projection depending on where its period falls relative to the edition's last historical point, and a projected cell sitting unnoticed in a historical panel quietly falsifies a backtest. Filter on period type and keep the forecast-completion date stamped on every pull.

Second, geography has a floor. The record resolves world, OPEC and non-OPEC aggregates plus United States national and PADD detail - country-by-country questions need a different record, and the honest pairing sums a country panel up to a matching aggregate before placing a forecast band beside it.

Third, units are part of the schema. Dollars per barrel, dollars per gallon, million barrels per day and percent share one table family, so normalize before computing any ratio, and treat the wholesale-versus-retail distinction as a real break in level rather than a rounding difference.

Fourth, revisions are features. Between the July and August 2026 editions the 2026 wholesale gasoline view moved from $2.75 to $2.91 a gallon - a 5.9 percent revision - and every future edition will revise again. Delivered histories keep the edition stamp so revision panels build as a query. Name the series, regions and periods you model when requesting the sample and it returns shaped to your book, dictionary holding steady - get a sample of this dataset and validate before any commitment.