2026 Edition · CC BY 4.0 · Permanent at /data/2026/

Download the data

Every number published in the atlas — county tier distributions, high+substantial job and payroll shares, exposure scores, ranks, imputation shares, and occupation-group detail for all 64 Colorado counties (2,860,276 UI-covered jobs, 2025 annual averages). The files are produced by the open pipeline; nothing is hand-edited.

Files

Files in the 2026 data bundle, with format and size.
File Contents Format Size
counties.csv Canonical flat file — one row per county, every published number. Columns documented below. CSV 12 KB
counties.geojson Simplified county geometry with core stats (high+substantial share, the former ≥50-cut share, band, score, rank, data quality) — the map layer. GeoJSON 65 KB
county/{fips}.json Per-county detail, including the SOC major-group tables behind each county page. Linked from every county page; example: county/08031.json (Denver). JSON × 64 411 KB total
state.json Statewide aggregates and validation statistics. JSON 2 KB
occupations.csv One row per detailed occupation — overlap tier (v1.2), exposure score, percentile, published Colorado employment and median wage (blank where BLS suppresses them), national employment. The data behind the occupation lookup; columns documented below. CSV 159 KB
occupations.json Same rows as occupations.csv, as JSON. JSON 379 KB
DATA_NOTES.md Auto-generated by the pipeline: source vintages, per-county imputation shares, validation results, caveats. Markdown 17 KB
README.md The bundle’s own documentation — file list and column schema, for readers of the raw directory. Markdown 6 KB
LICENSE CC BY 4.0 legal code. Text 18 KB

counties.csv columns

Column schema for counties.csv.
Column Meaning
fips 5-digit county FIPS code
name County name (Census spelling)
slug URL slug — the county page lives at /county/{slug}/
total_emp 2025 annual average UI-covered employment (QCEW)
exposure_score Employment-weighted mean AI exposure score, 0–100
rank Rank on exposure_score, 1 = highest (computed before rounding). Note this orders the mean score, while band is cut on high_substantial_share_pct — the two can disagree
high_substantial_share_pct v1.2 HEADLINE: % of jobs in occupations with high or substantial task overlap (exposure score ≥ 45 of 100)
high_substantial_jobs v1.2: count of jobs behind the headline share (share × total_emp, at full precision)
payroll_high_substantial_share_pct v1.2: % of estimated payroll in high- or substantial-overlap occupations
tier_high_pct, tier_high_jobs v1.2: share and count of jobs in high-overlap occupations (score ≥ 60 — most tasks overlap)
tier_substantial_pct, tier_substantial_jobs v1.2: share and count in substantial-overlap occupations (45 ≤ score < 60 — a large minority of tasks)
tier_some_pct, tier_some_jobs v1.2: share and count in some-overlap occupations (30 ≤ score < 45)
tier_little_pct, tier_little_jobs v1.2: share and count in little-overlap occupations (score < 30)
band v1.2: 1–5 share quintile by county count on the headline share, 1 = well above the typical Colorado county
band_label the band written out, anchored to the typical (median) county
rank_suppressed true where more than 30% of employment is imputed — the site withholds the ordinal for these counties and shows the band instead; the rank column is still populated here
high_exposure_share_pct Alternative cut (v1.1): % of jobs under the former binary ≥50 threshold — retained for continuity, not the headline
high_exposure_jobs Alternative cut (v1.1): count of jobs under the ≥50 threshold
payroll_high_exposure_share_pct Alternative cut (v1.1): % of estimated payroll under the ≥50 threshold
emp_imputed_pct % of employment imputed due to QCEW suppression
emp_fallback_pct % of employment routed via fallback staffing patterns
data_quality full / partial_imputed / heavily_imputed (see DATA_NOTES.md)
population Context: county population, 2020–2024 ACS 5-year estimates
population_moe Context: population margin of error (blank where the Census estimate is controlled — most counties)
median_earnings Context: median annual earnings for workers living in the county, 2020–2024 ACS, dollars
median_earnings_moe Context: earnings margin of error (90% confidence), dollars

Rounding: scores and percentage shares to 1 decimal, employment to integers. Ranks are computed at full precision, so two counties can share a rounded score with different ranks. Major-group employment in the per-county JSON is rounded independently and need not sum exactly to total_emp. The four context columns are residence-based ACS estimates that describe a county's people without entering any exposure computation, and are the only columns that may be blank. Exposure is not a job-loss forecast — see DATA_NOTES.md.

occupations.csv columns

One row per detailed SOC 2018 occupation (military excluded) — the data behind the occupation lookup. Entirely published data joined to published scores, with no imputation: where BLS suppresses a Colorado value the cell is blank.

Column schema for occupations.csv.
Column Meaning
occ_code SOC 2018 detailed occupation code
slug URL slug — the occupation page lives at /occupation/{slug}/
title Published OEWS occupation title, verbatim
major_group SOC major group code (XX-0000)
group_title SOC major group title
score AI exposure score, 0–100 (Eloundou human β × 100)
score_source direct, or imputed_major_group where no task-level rating is published
tier v1.2: high / substantial / some / little — overlap tier, cut at 60/45/30 on the score
tier_label v1.2: what the tier means, written out ("most tasks overlap with current AI capability")
zero_beta v1.2: true where the published rating is exactly zero (55 occupations) — site surfaces show "below measurable task overlap" rather than 0.0; the exact value is here
percentile % of occupations with a strictly lower score (0–100, unweighted)
co_emp Published OEWS Colorado employment (blank where BLS suppresses it)
co_median_wage Published OEWS Colorado median annual wage, dollars (blank where unpublished)
national_emp Published OEWS national employment

Wage flags follow the pipeline's conventions: top-coded medians (#) resolve to the published threshold, a conservative floor; hourly-only occupations are annualized as the published hourly median × 2080. Details in the bundle README.

License & citation

The data is licensed Creative Commons Attribution 4.0 (CC BY 4.0): use it for anything, including commercially, with attribution. The pipeline code is open source under the MIT license. Upstream sources carry their own terms, listed in the methodology source table; the Felten AIOE raw values are unlicensed and are never redistributed here.

Requested citation

Colorado AI Exposure Atlas


Colorado AI Exposure Atlas, 2026 edition. Christopher Martin. https://coloradoaiexposureatlas.com/

2026 Edition · Employment data 2025 · Compiled by Christopher Martin