Is the Baker Hughes rig count data free? The census behind the number, delivered as rows

Depth that surprises people who have only ever seen the headline chart. The current verified North America workbook runs to roughly 8 MB with six tabs - summary, breakdown, yearly, quarterly, monthly and weekly - and the weekly tab alone holds more than 110,000 rows back to January 2000. Each row carries the full dimensional context of one count week: country, county, basin, a Gulf of Mexico flag, drill-for target, land / inland waters / offshore setting, state or province, trajectory, year, month, publish date and the count value.

That dimensionality is the entire point. Because every row arrives pre-dimensioned, a Permian-only panel, a Montney-only panel, an offshore-versus-land split or a horizontal-versus-vertical mix is a WHERE clause instead of a stitching project. The DrillFor column deserves special mention: it records the operator's declared commodity target at permit issuance, which means drilling intent is visible before any barrel flows - the gas-directed versus oil-directed split traders quote every Friday is a single filter on this column, defined once as horizontal when inclination exceeds 80 degrees.

The archive shelf behind the current report runs deeper than most analyses ever use: North America archive series reaching back to 1975-era masters, worldwide archives covering 1975 through 2024, Rigs-by-State tables for 2000 through 2024, US annual averages by state for 1987 through 2016, and a discontinued workover-rig series spanning 1999 through 2007. The worldwide half adds its own texture: monthly counts by region and country in the current format from 2011 onward, with Saudi Arabia reporting on an operating-rigs basis rather than the active basis used everywhere else.

Three captured rows from the January 2024 Canada detail, exactly as the dictionary shapes them:

# one count week, one dimension combination - Canada, January 2024
country           : CANADA
state_province    : ALBERTA
county            : Foothills Front
basin             : Other
gulf_of_mexico    : No
drill_for         : Gas
location          : Land
trajectory        : Horizontal
year              : 2024
month             : 1
rig_count_value   : 36

# same county, same week, drilling for oil instead
drill_for         : Oil
rig_count_value   : 18

# the next county over, gas-directed
county            : Central Alberta
drill_for         : Gas
rig_count_value   : 2

Read the anatomy rather than the values. One row resolves to a single county-week-target cell - Foothills Front, Alberta, January 2024, gas-directed - and hands over its count (36) with the full dimensional context attached. The second row is the same county and week with the target flipped to oil, printing 18: two commodities sharing one geography means the split is a filter, not a separate dataset. The third widens one county over, where 2 gas-directed rigs were all that turned that week.

Twelve verified columns do the work, and four of them decide whether a model survives contact with reality.

Geography stacks three deep: State/Province narrows to County, the finest split the weekly detail carries, while Basin reads Other wherever a rig sits outside a named play - worth knowing before any per-basin normalization, because Other is a bucket, not a basin. The Gulf of Mexico flag peels deepwater activity out of the general offshore count, which keeps short-cycle shale reads from being contaminated by five-year deepwater projects.

Intent and shape ride beside geography. DrillFor captures declared target at permit issuance; Location splits land against inland waters and offshore; Trajectory marks vertical, directional or horizontal with the 80-degree definition stated outright, so a horizontal count means the same thing in 2004 and 2024.

Time arrives twice, and that is deliberate. Year and Month timestamp the count week as calendar fields, while the publish date serializes as an Excel serial number keyed to the Friday noon Central release. Joining the census to a price panel on dates requires converting that serial once - handled upstream in a Datadory delivery, discovered mid-model otherwise.

One reading rule prevents most cross-vendor confusion before it starts: anchor any panel to one census. Count definitions differ between compilers, and mixing series produces phantom inflections - Saudi Arabia's operating-rigs basis inside an otherwise active-rigs panel is the classic case.

Geography spans the United States end to end by state, county, basin and Gulf of Mexico flag; Canada by province; and worldwide monthly counts by region and country. The worldwide series carries a structural exclusion list that belongs in every global panel plan: Russia, the Caspian, Iran, Sudan, Cuba, North Korea and onshore China sit outside the census, Syria since February 2012, Iraq between September 1990 and May 2012. Those gaps are design decisions, not missing files, so plan for them rather than discovering them mid-analysis.

Temporal depth splits by hemisphere of the record. North America: weekly archive series back to 1975-era masters, with the full-dimensional weekly detail table running January 2000 to present. Worldwide: master archives from 1975, current-format monthly reports from 2011. Either way, the current week always lands inside a near-five-decade arc of booms, busts and recoveries - long enough to carry a full drilling cycle in-sample rather than treating every downturn as unprecedented.

Granularity stops at the segment level, and saying so plainly is more useful than pretending otherwise. Rows resolve by week and dimension - there is no operator identity, no wellhead, no lease underneath. Asset-level questions belong to the regulator registries: the Texas Railroad Commission data sets for every well ever permitted in Texas, the BSEE Well API for all 57,482 Outer Continental Shelf boreholes, or the North Dakota Industrial Commission data whose live rig list names each Bakken rig, operator and next location.

The census answers how many rigs turned. Four sibling records answer what surrounds them:

• EIA Crude Oil and Natural Gas Drilling Activity (Rotary Rigs) is the longest continuous view of American drilling effort: seven series from 1949 - total, onshore and offshore rotary rigs plus crude-oil-target versus natural-gas-target counts from 1973 - opening at 2,017 rigs and closing at 599 in 2024. EIA compiles it from the Baker Hughes weekly count itself, so the two agree by construction; the pairing buys you seventy-five years instead of thirty-five. One convention matters: annual rows are dated June 30 and average 52 or 53 reporting weeks, not calendar years.
• EIA Drilling Productivity Report converts rig effort into economics: monthly rig counts, new-well production per rig and drilled-but-uncompleted inventories for Anadarko, Appalachia, Bakken, Eagle Ford, Haynesville, Niobrara and Permian, one consistent regional key from January 2007 through June 2024, continued in the Short-Term Energy Outlook after May 2024.
• AER Alberta Energy Outlook (ST98) extends the frame past the border and forward in time: Alberta well counts by PSAC area and formation from 1999, with a rolling ten-year forecast reaching 2034 - the one trajectory in the slice no US source replicates.
• OilPriceAPI Drilling Intelligence - Well Permits adds the leading indicator: 2.19 million-plus permits aggregated across 27 state regulators, with weekly US, Canada and international rig counts riding the same feed.

Stacked deliberately, they cover the full chain - permit, rig, well, production per rig, forecast - and the first table below lines them up side by side.

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

1. Quant and fundamental investors treat the weekly count as the market's coincident activity pulse - the gas-versus-oil split signals which price deck operators are answering, and the multi-decade runway supports honest backtests of activity-to-production lags. Their page: oil & gas drilling for investors and quants; their workflow: quant backtesting.
2. Data scientists and ML engineers get activity features on a stable dimensional grid that joins to price panels on dates without entity resolution - the county-and-basin grain turns feature engineering into aggregation. Their workflow ranks models in ml-model-training.
3. Market researchers and consultants anchor drilling-activity chapters on the citation everyone already recognizes, then size the service economy around the rigs with permit and operator rankings. See market sizing.
4. Journalists, academics and students quote a published census with a stable definition maintained across five decades - and cite it with receipts at county grain. Related standard: citation-grade research.
5. Oilfield-service and equipment teams read regional and method splits as a demand map - horizontal-directed weeks concentrate in shale plays and pull different pressure-pumping demand than vertical work, weeks before order books confirm it.

The constant across all five: because every row carries its full dimensional context, aggregation upward is arithmetic rather than reconstruction.

That last clause matters more here than in most slices. The census publishes weekly, but nobody consumes it weekly-only: a monitoring loop wants Friday's print within the hour, a backtest wants the whole 110,000-row detail table once, and a dashboard wants both under one schema. Delivered rows make those three jobs the same table; self-assembled pipelines make them three integrations.

Because the number was never the hard part - the seams are. Excel serial publish dates sitting beside calendar year-month pairs. A Basin column whose Other bucket silently absorbs unassigned rigs. An exclusion list governing exactly which countries a worldwide panel may contain. Two commodities hiding in one geography until the DrillFor filter is applied. Vendor definitions that differ enough to manufacture inflections no operator ever drilled.

Each is survivable once. None is fun to re-solve in every new notebook. Datadory handles them upstream: dates typed and converted, buckets documented, exclusions carried as flags beside the rows they affect, and the whole dictionary held steady between the sample and the standing feed - so anything prototyped on the sample survives delivery intact.

Name the geographies and segments when you request a sample and judge the columns, not the promise. It arrives with the field dictionary and coverage statement attached, yours to keep either way.

Start with the record pages: the Baker Hughes Rig Count dataset page holds the full twelve-field dictionary and sample rows, and Baker Hughes Rig Counts via ICE Data API documents the commercially distributed segmentation since 1990.

For the whole slice in one place, the oil & gas drilling industry hub indexes all 27 pooled datasets with field dictionaries and sample rows, and best oil & gas drilling datasets scores them on one rubric. When you're ready to build, request a sample cut to the basins on your desk this quarter - the schema in the sample is the schema you ship against.