Corrections

Data corrections & errata

Figures this dataset published and later found to be wrong — what each one was, what caused it, and when it was corrected. Newest first.

This page is the living record. Archived snapshots of the dataset are fixed at the moment they were deposited and cannot be edited, which is deliberate — a caveat written for one snapshot describes that snapshot. Anything learned afterwards is recorded here instead. Back to the downloads.

Corrected 2026-08-13

Michigan oil under-collected by up to 13%

Michigan, 2020-01 to 2026-05 — statewide totals, all 63 counties, and the county CSVs

What was wrong: Statewide Michigan oil was short by 2.3% in 2020, rising to 13.2% by 2023 and holding there. Because the missing volume was concentrated in the counties that were growing, the series showed Michigan oil in decline while the state regulator showed it rising from 4.19 to 4.81 million barrels across 2020–2023.
Cause: The fetcher queried Michigan’s Data Miner once per county, filtering server-side on a county field. For newer, high-numbered production units that filter fails to resolve the county at all, so those units matched none of the 83 queries and were counted in none. The omitted units were the state’s two growth stories — Trenton–Black River horizontals in Kalamazoo and Calhoun, and a CO2 flood in Otsego.
Correction: Replaced the per-county sweep with the regulator’s bulk production and well downloads, which need no county filter. Michigan now reconciles to within 0.5% of the independent EIA series for 2022–2024, where it previously sat 13–15% below. A new check refuses to publish when more than 0.5% of statewide volume is unattributed to any county.

Corrected 2026-08-13

California missing six months of published production

California, 2026-01 to 2026-05 — absent entirely

What was wrong: The California series ended at 2025-12 while the regulator had already published through 2026-05 — roughly 36.7 million barrels of oil that existed upstream and never appeared here. Every other state was current to within three months.
Cause: The fetcher assumed CalGEM publishes only a completed prior-year file each March. It also maintains a live current-year file that grows monthly, and nothing was looking for it.
Correction: The fetcher now asks the endpoint which years exist rather than assuming, and falls back to the prior year if the probe fails. California now runs through 2026-05, in line with the other states.

Corrected 2026-08-04

Kansas served roughly twice the real production

Kansas, all counties and all periods, until 2026-08-04

What was wrong: Kansas county production read approximately 2× reality. Any total, chart or download taken from the Kansas series before this date is overstated by roughly a factor of two.
Cause: The source file carries a “State Total” row alongside the county rows. It was ingested as if it were a 95th county, so the statewide figure was added to the sum of the counties that comprised it.
Correction: The total row is excluded from county ingestion. Kansas now reconciles against the independent EIA series.

Corrected 2026-08-04

Illinois presented derived figures as measured

Illinois, all counties — the labelling, not the volumes

What was wrong: Illinois county monthly figures are allocated from state totals rather than reported per county, but the archive carried a flag saying otherwise and the site rendered derived barrels-per-day as if measured. The numbers were what they always were; the claim about them was wrong.
Cause: The synthetic flag was set false on a series that is wholly derived.
Correction: Illinois is flagged as derived at every level — envelope, county and row — and every generated CSV carries a “Derived” header line. The deposit README names it explicitly.

Corrected 2026-08-04

Colorado months shifted for sparse counties

Colorado, counties with intermittent reporting

What was wrong: Production was attributed to the wrong month in counties that do not report every month. Annual and statewide totals were unaffected, which is why it survived so long — only the month-level shape was wrong.
Cause: The parser advanced its month cursor per row rather than per reported period, so a gap displaced everything after it.
Correction: The period is read from each row. A month-level reconciliation against EIA now runs weekly and would catch a recurrence.

Corrected 2026-08-04

Wyoming coverage overstated by 35 years

Wyoming — the stated start date, not the volumes

What was wrong: The archive advertised a start period 35 years earlier than the data it actually held. Anyone reading the coverage claim would have expected history that was not there.
Cause: The start period was written from a constant rather than measured from the rows present.
Correction: Coverage dates are computed from the data at write time across every state.

What does not appear here

Defects caught before they reached anyone are not errata. A guard that blocks a bad refresh has prevented a correction rather than caused one, and listing those would make this page a changelog instead of a record of what readers were actually told.

Known differences from other published sources are a separate matter and are documented rather than corrected — where this dataset and the U.S. Energy Information Administration disagree and nobody has established which is right, that stays on the record as an open question. See the methodology and the divergence document shipped with each archived snapshot.