Commodity Flow Survey (CFS) 2022

One survey stands underneath nearly everything America knows about its own freight: the Commodity Flow Survey, run jointly by the U.S. Census Bureau and the Bureau of Transportation Statistics, measuring domestic shipments by US establishments every five years. It is the primary microdata source for ground cargo flows and the foundation of the Freight Analysis Framework.

The 2022 cycle arrived in stages. Final estimates landed 20 May 2025, the summary tables followed on 26 June 2025, the survey publication came 20 January 2026, and the shipment-level Public Use Microdata Sample closed it out on 28 January 2026. That last piece is the prize: identifiers running 00000001 through 37576546 - 37,576,546 possible shipments, each carrying where a load started, where it ended, what it was, how it traveled, what it was worth and how far it went. Datadory scores it 10 out of 10 on its quality rubric - a mark held by just 145 of the 1,744 cataloged datasets against a 7.81 average. Get a sample of this dataset and see the rows before anything else.

Flat, wide, keyed - one row per sampled shipment. Straight off the verified 2022 dictionary:

SHIPMT_ID     : 00000001
ORIG_STATE    : 06        DEST_STATE : 48       # California -> Texas
SCTG          : 35        MODE      : 02       # truck move
SHIPMT_VALUE  : $125,000
SHIPMT_WGHT   : 42,000 lbs      SHIPMT_DIST_GC : 1,745 mi

Read it as one truckload priced three ways: dollars, pounds and mile-weighted effort. The second row shape shows the classifier half of the record:

SHIPMT_ID     : 01234567
ORIG_CFS_AREA : 31080     DEST_MA   : 19100    # metro-to-metro lane
SECTOR        : 31-33     TEMP_CNTRL_YN : N    HAZMAT : 0
EXPORT_YN     : N         WGT_FACTOR : 12.4

That last field is the quiet one doing the heavy lifting. WGT_FACTOR is the tabulation weight that inflates the sample into population estimates - skip it and your national totals describe only the surveyed fraction of American freight. With it, summed weighted SHIPMT_VALUE reproduces the headline figures the summary tables publish, which is exactly the cross-check that keeps a model honest.

Eighteen documented variables, every one defined and worked into an example during verification. The table below carries the ten that do most of the analytical work: three nested origin-destination pairs (state, metro area and the survey's own CFS area), the shipper's industry SECTOR, the SCTG commodity code, the transportation MODE - which distinguishes for-hire truck from private truck, a distinction most freight datasets flatten away - plus dollar value, pound weight and great-circle miles.

The remaining eight ride under additional fields on request: the temperature-control flag, the export pair and country, the hazmat class and the tabulation weight. Field definitions are marked verified, meeting the standard held by 1,495 of the 1,744 datasets (85.7%) in the Datadory catalog.

Geography - three nested cuts, twice over. Every record holds origin and destination as a FIPS state pair, a metropolitan-area pair and a CFS-area pair, so a corridor reads at whatever resolution the question needs. Suppression is coded rather than dropped - 00 for a withheld origin state, 00000 for a withheld CFS area - which is the correct way to handle confidentiality but the wrong thing to ignore in a row count.

Temporal - the 2022 survey year, delivered across a staged release train: final estimates May 2025, summary tables June 2025, publication and PUMS microdata in January 2026. The survey runs every five years, and the 2017 and 2012 cycles are available alongside, which makes two-cycle comparisons straightforward and continuous annual series impossible - pair it with a monthly indicator when between-cycle freshness matters.

Granularity - shipment-level microdata, one row per shipment, plus roughly 30 pre-tabulated summary tables cut by commodity, mode, distance band and geography for when the question stops needing row-level detail. The microdata file runs large - hundreds of megabytes compressed - while a single summary table fits in a spreadsheet; Datadory ships either shape, cut to your commodities and lanes.

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

The cadence follows your pipeline, not the other way around. Rows arrive flattened and typed, keyed on the shipment identifier, with the mode, SCTG and sector code sheets attached so nothing lands as an unexplained integer - joining a delivery against carrier telematics, facility locations or a rate table is a join statement rather than a documentation archaeology project. A sample cut to your commodities, lanes and geographies comes first either way.

• Market sizing - weight SHIPMT_VALUE to population by commodity, mode or distance band and the US domestic freight market gets a denominator with a federal pedigree. See market researchers x cargo ground transportation.
• Network planning - rank origin-destination pairs by value, weight or ton-miles before siting distribution capacity; metro and CFS-area endpoints make the ranking lane-specific rather than state-vague.
• Demand modeling - millions of shipment records with value, weight, distance and commodity feed lane-choice and freight-demand models directly; part of the shelf serving supply-chain mapping.
• Investment backtests - the for-hire versus private-truck split in MODE lets quants test tonnage proxies against carrier fundamentals across cycles; see investors & quants x cargo ground transportation.
• Cold-chain and hazmat studies - the temperature-control and hazmat flags isolate refrigerated and dangerous-goods flows by commodity and corridor without a second source.
• Reporting and coursework - the official citation-grade measure of what moves across America, suppression rules included; see journalists & academics x cargo ground transportation.

Market Researchers & Consultants get the federal benchmark under every freight TAM model (market researchers view). Data Scientists & ML Engineers get a clean keyed microdata set that trains demand models without scraping work (data scientists view). Investors & Quant Researchers get cycle-over-cycle trucking-demand proxies they can test against carrier revenues (investors view). Journalists, Academics & Students get quotable national freight figures with the methodology already written (journalists view). Developers & Data-Product Builders get a stable schema to seed freight-lane lookup features from (developers view).

Stated plainly, because they shape the analysis:

• Five-year means five-year. This is a benchmark, not a monthly indicator; intra-cycle movement needs a companion series, and treating a quinquennial snapshot as a leading signal misuses it.
• Suppression is coded, not absent. Values of 00 and 00000 mark withheld geography; summing past them silently biases fine-grained regional work toward the areas the survey could publish.
• Weights are not optional. WGT_FACTOR turns a sample into the economy; skipping it produces numbers that look precise and describe only the surveyed fraction.
• Distance is great-circle. SHIPMT_DIST_GC measures straight lines, not road miles, so route-length calculations need a routing layer on top.
• Code labels live in the companion sheets. Mode and sector arrive as codes; read them against the full dictionary - shipped with every delivery - before assuming any particular value.

Because the raw release measures in the hundreds of megabytes, splits its dictionary into a separate workbook, and expects you to reconcile staged releases that arrived eight months apart. Datadory normalizes the microdata and the summary tables into the single field dictionary above, resolves code sheets onto every row, flags suppression codes explicitly instead of leaving them to be discovered downstream, and schedules recurring deliveries keyed on the shipment identifier. Browse the shelf on the cargo & ground transportation data hub or the best cargo ground transportation datasets, then get a sample cut to your lanes.

Neighbors that bracket the questions this survey leaves open. Freight Analysis Framework Version 5 (FAF5) republishes these flows as an origin-destination matrix across 132 domestic regions with forecasts to 2050 - the head-to-head runs in vs Freight Analysis Framework Version 5 (FAF5). FHWA's Freight Performance Measurement Program adds highway travel-time reliability between the endpoints. ATA Economics' truck tonnage measures track the month-to-month pulse between survey years. Eurostat and UK road-freight statistics extend the same questions across the Atlantic, and Data.gov's catalog reaches everything else published on US freight. Together: the microdata, the modeled matrix, the highway reality, the monthly pulse and the international comparison.