Country registry
All documented registry entries, not only task countries
Open research data
The observations behind the figures.
35 curated evidence tables, with previews, field definitions and direct downloads. Every download contains its complete frozen table.
Start with the native occupation data. Country exports preserve original shares and publication status. Broader task and mapping tables answer different questions. Choose the right measure.
35 datasets · 581 documented fields
Global availability
Missing source coverage is not zero occupation coverage
Source country period inventory
Displayed-zero task rows are retained separately; a period without positive cells does not acquire an occupation panel
Country coverage
Source cells counted once before occupation expansion; numeric zero rows separate
Global employment latest
Source year/detail/age/flags retained; national classifications not silently relabelled
Labour availability
All250 registry entries; employment availability differs from task join feasibility
PIAAC country source inventory
England/Flanders scopes retained; no respondent records
Country usage
Overall source fields; absent country metrics remain absent
Direct occupation source
Only original numeric rows, including rounded zeros
O*NET occupation panel
Absent raw values missing; zero published mass is not zero actual usage
ESCO occupation panel
Semantic mean, not conversation allocation
ISCO occupation panel
ISCO1/2 aggregate directly from ISCO4
Facet comparison
Task and occupation mass counted before expansion
Volume regression
Selected publication sample; no causal interpretation
US bls employment
National Employment Matrix includes self-employment; projections excluded
US native soc panel
Exact codes only; unmatched BLS codes remain missing
US native weighted summary
Official total includes unmatched job groups
US measure validation
Cross-construct and cross-period comparison with March exposure
US validation pairs
All target occupations, no European crosswalk
Mapped weighted summary
UK ONS lexical route; US excluded in favour of native BLS
Geography support
Post-review thresholds; missing May facet grade D
PIAAC across occupation summary
All finite coefficients included regardless of sign/review status
PIAAC detailed country results
Retains full outcome/profile variants
Occupation mapping correlations
Direct-share descriptors; all typed reference
Mapping correlation distribution
Common occupation counts and country counts are distinct
Occupation comparator correlations
Aggregate correlations only; restricted source vectors omitted
Comparator distribution
Unlike constructs; no calibration or cardinal subtraction
Allocated occupation usage
Conditional allocation of published usage. equal_all splits each source evenly across distinct finest known ISCO destinations and then sums; UK continues through an equal reverse ONS bridge. Three-digit nodes remain unmatched at four digits. Sensitivity scenarios and marginal mapping bounds are not confidence intervals. Shares retain the country consumer denominator. Employment ratios use full official totals and separate vintages; not adoption rates.
Occupation allocation weights
Each source row sums to one, including UNMATCHED. Independent of country employment and outcomes. UK lexical_all uses reverse title counts, not worker transition probabilities.
Allocation accounting checks
Allocated plus unmatched equals original published occupation mass. Unpublished residual is separate. Arithmetic tolerance 1e-9 percentage points.
April–May occupation comparison
Same source release and fixed allocation weights in both months. Original denominators retained; published source support may differ. Earlier weekly windows are excluded.
UK usage, employment and pay panel
APS 2021–2025 calendar-year employment; ASHE April nominal hourly pay, excluding overtime. Different populations and reference periods retained. Outcomes predate April–May 2026 usage; this is not a forecast test. Choose one AI month/scenario before analysis; repeated outcomes are not independent observations.
UK usage and recruitment-advert panel
ONS/Textkernel new adverts 2017–July 2026, not successful hires. Source gaps, suppression and period notices retained. July is the only full month after predictor release in this snapshot and has quality warnings. No predictive or causal effect is estimated.
European usage and employment panel
Eurostat annual employment 2020 onward joined to fixed all-link equal allocations for April/May 2026. Break flags, population, dates and national source metadata retained. Not a causal estimate or a forecast; do not treat the two AI-month copies as independent outcomes.
Multilingual occupation reference
Official ESCO v1.1.2 labels and available descriptions. Corresponding languages share the same source URI and ISCO parent. This is a reference set for evaluating label retrieval, not a validated classifier of conversations, local work practices or users’ jobs. No model accuracy is claimed.
Reproducible and attributable
Source observations: April and May 2026. Allocations and research extensions are documented in the current file manifest. The curated web export preserves source terms and source precision.