EIA bulk download petroleum archives: the federal fuels catalog, delivered as rows

Reproducibility is the whole argument. Each version is a frozen capture that ends at its assembly date, addressable by number, so a feature built against 81.0.0 returns identical figures months later - something a live feed cannot promise once the agency revises a week. The manifest ships a hash per resource, which reduces integrity checking to a one-line diff, and a pinned rebuild reproduces bit-for-bit.

Versioning also moves fast enough to matter: version 54.0.0, published January 25, 2026, held 26 files totalling about 1.46 GB; version 81.0.0, seven months later, holds 30 files near 1.6 GB. Chasing that by hand is a part-time job. Delivered through Datadory, the version pin travels with every extract - a dashboard, a model's training table and a board deck can cite the same capture - while newer captures slot in as their own numbered versions, never as edits underneath you. Currency becomes a settings conversation instead of a re-integration project.

Whole-corpus delivery is the norm here rather than the exception: roughly 725 of the 1,744 records in Datadory's catalog arrive as complete file sets rather than per-series requests. The difference is who babysits the loop.

Weekly inventories lead the list. A representative series, PET.WO6ST_R20_1.W, tracks Midwest (PADD 2) ending stocks of conventional CBOB gasoline blending components in thousand barrels; its history in the current capture runs from April 9, 2004 to January 16, 2026, with the agency's most recent revision stamped January 22, 2026 - about a week between an observation week and the final print, which is the lag to budget for when quoting the freshest week.

Its three newest observations read 40,412, 38,357 and 35,617 thousand barrels for the weeks ending January 16, January 9 and January 2, 2026 - a 4.8-million-barrel two-week build visible in three lines of text. That structure generalizes across the file: weekly PADD, lower-48 and state stocks plus import series carry observation histories observed from April 2004 forward, while the imports companion adds city, port and refinery detail segmented by crude grade and country of origin. Traders tracking inventory cycles and quants assembling positioning features end up reading the same lines.

One record per EIA series, exactly as the bulk text returns it - this shape was verified against the actual file during catalog research:

# One record per EIA series, exactly as the bulk text carries it
series_id       PET.WO6ST_R20_1.W
name            Midwest (PADD 2) Ending Stocks of Conventional CBOB
                Gasoline Blending Components, Weekly
units           Thousand Barrels
frequency       W (weekly)
geography       USA-IA+USA-IL+USA-IN+USA-KS+USA-KY+USA-MI+USA-MN+
                USA-MO+USA-ND+USA-NE+USA-OH+USA-OK+USA-SD+USA-TN+USA-WI
start           20040409          end   20260116
last_updated    2026-01-22T21:12:35-05:00
data (latest)   20260116  40412   |  20260109  38357   |  20260102  35617

Two things fall straight out of that shape. The geography spec enumerates all fifteen states summed into PADD 2 and rides the row itself, so a dashboard at district level and a state-by-state study share one parser - climbing or descending the geographic ladder changes no schema. And because the revision stamp travels with every series, vintage handling becomes a filter downstream rather than a footnote nobody reads.

• Geography: United States at national, PADD and state level for stocks; imports descend to city, port and refinery grain by crude grade and country of origin; the international sets extend coverage to individual countries.
• Temporal: weekly stocks observed from April 2004 forward, Annual Energy Outlook editions reaching back to 2014, and every version ending cleanly at its capture date - August 16, 2026 for 81.0.0.
• Granularity: individual EIA series, tens of thousands in the petroleum file alone, each carrying its complete observation history rather than a windowed excerpt.

Two edges deserve planning rather than surprise. First, revision lag: expect roughly a week between an observation week and the agency's final print, stamped in the record itself. Second, the capture edge: anything after a version's assembly date arrives in the next numbered version, which lands on your schedule once delivery is configured - never as a silent patch to history you already quoted.

Ranked by how directly the weekly, geographically-split grain answers the day job:

1. Investors and quant researchers. Weekly PADD and state stocks are the supply-demand balance gauge behind inventory-cycle positioning; investors and quants use cases shows where the series slot into screens and backtests.
2. Data scientists and ML engineers. Pin training inputs to a numbered version so a rebuild reproduces bit-for-bit; the pattern is laid out in data scientists use cases.
3. Developers and builders. One scheduled versioned object replaces per-series request loops and pagination arithmetic - plumbing that stops being maintained by hand.
4. Supply and trading analysts. Imports sliced by port, refinery, crude grade and origin turn origin-shift questions - which Gulf Coast grades came from where this quarter - into queries.
5. Journalists and academics. Cite one named version and the figures stay checkable indefinitely, which is precisely the property citation-grade work depends on.

Persona tagging agrees: data scientists plus developers score four on relevance, journalists and academics two, investors and market researchers one.

This archive answers how much; the rest of the shelf answers where. OGIM — Oil and Gas Infrastructure Mapping database (EDF) scores 9/10 and maps the physical estate: 6.7 million harmonized features across 152 countries - 4.54 million wells, 1.86 million pipeline segments spanning more than 1.2 million kilometres, 3,661 petroleum terminals, 692 refineries and 547 LNG facilities. The head-to-head comparison scores that volumes-versus-locations trade-off line by line, and the strongest desks run both.

Closer to the tank, HIFLD Petroleum Terminals maps all 2,302 operable US bulk terminals with fifty attributes apiece, TGS US Pipelines contributes 731,388 segments carrying operator, status, diameter, capacity and flow direction, OpenStreetMap adds the only realtime layer - roughly 867,000 tagged storage tanks - and TankTerminals.com goes commercial-global with capacity fact sheets across 13,000-plus facilities in 2,420 ports. Join key throughout: geography plus product plus period.

Start with the oil & gas storage & transportation data guide for the full eight-record pool, or the industry hub for how the volumes layer sits beside the asset layers. For the gas-side twin of this workflow, see how EIA weekly storage report history arrives as one continuous regional panel. The PUDL Raw EIA Bulk API Data dataset page documents sample rows, the field dictionary and coverage in full.

Request a sample scoped to the packages, series families and era you actually need - real rows come back with the manifest attached before anything recurring switches on.