Validation

Validation is claim-specific. Reproducing arithmetic, comparing occupational constructs, testing a classification bridge and assessing agreement with survey activities address different questions. Each result below retains its sample and target claim.

Native US construct comparison

The United States permits comparison before an ESCO or ISCO crosswalk. Direct occupation shares are summed to native six-digit SOC groups; task descriptors are averaged within those groups, with source rounding ties preserved.

The direct shares align more closely in this comparison, but none of the coefficients establishes equivalence. Monthly consumer measures and the March exposure construction differ in period, sample and definition. Exposure combines work-related use, capability, automation and task-time aggregation. See US measure validation and its download.

Exposure reconstruction

Let $C^{\mathrm{work}}_t$ denote work-related consumer counts for task $t$, $A_t$ global API counts, $q_t$ the consumer automation share and $\beta_t$ the capability score. The published method first defines

and, for $W_t>0$, the automation factor

The gated task value is

with $r_t=0$ when $W_t=0$, without evaluating the zero denominator. For occupation-specific task-time weights $w_{ot}$, exposure is

Both sums range over the occupation's full task set. Tasks that fail either gate contribute zero to the numerator but remain in the time-weight denominator; weights are not renormalized over eligible tasks. This algebra shows why reproducing the final score requires intermediate counts, automation shares, gates, grouping and time weights rather than occupation shares alone.

The independent public-input proxy achieves a Spearman correlation of 0.89465 across 756 occupations and shares 6 occupations with the published top ten. It does not meet the prespecified criteria of at least 0.95 and at least 8 common top-ten occupations. Equal-weight aggregation of the publisher's task scores reaches 0.87154, but that downstream diagnostic already embeds publisher outputs.

Original time fractions, work-share imputation, similar-task grouping, employment allocation and uncensored intermediates are not all recoverable. The evidence does not establish that any one missing input explains the disagreement. The direct occupation dataset remains valid as a record of published shares; it is not presented as a reconstruction of exposure.

Native employment join

The BLS National Employment Matrix contains 831 detailed line items and 170,280,800 base-year jobs in 2025. Exact code matching supports 772 detailed groups and 92.53% of the official total. This verifies the join and denominator. It does not show that platform users represent those jobs. The Matrix counts jobs, including unincorporated self-employment, rather than individual workers. See US bls employment and BLS definitions.

PIAAC content comparison

The OECD Survey of Adult Skills records the frequency of work activities. An explicitly interpretive bridge relates survey items to seven O*NET activity domains. At ISCO two-digit level, the primary comparison ranks occupations by the survey share reporting an activity at least weekly and by the corresponding share of catalogue tasks.

Agreement varies by domain and becomes uneven at finer occupational detail; weak and negative coefficients remain in the outputs. Each public activity cell requires at least 30 valid respondents. Survey weights enter the activity shares, but these rank summaries do not claim replicate-weight confidence intervals. Survey geographies such as England and Flanders retain their actual scope. See PIAAC across occupation summary and PIAAC detailed country results.

Frequency is not time allocation, and catalogue shares are not representative AI adoption. PIAAC therefore supplies a partial work-content comparison rather than validation of task-time weights or national exposure.

Classification sensitivity

European mapping alternatives are compared on pairwise common occupational support. A strong rank correlation on that subset can coexist with missing destinations or large changes for particular occupations. Exact, exact-plus-narrow, all-typed and untyped routes are sensitivity specifications; they are not statistical confidence bounds.

The study also compares occupation descriptors with ILO, OECD, Eloundou and Felten capability measures. Score direction and paired sample sizes are retained, including negative coefficients. Theoretical feasibility and observed platform activity are related but distinct constructs. See Mapping correlation distribution and Comparator distribution.

Excluded time-weight experiment

A local-model experiment produced task-hour estimates that failed basic plausibility checks, including an extreme task value and occupation totals above the number of hours in a week. The weights and resulting correlations are excluded from substantive conclusions; clipping or normalizing invalid outputs would not validate them.

The prospective replacement protocol requires three blinded independent draws, exact task identifiers, finite nonnegative allocations, an explicit total-time constraint and whole-draw rejection rules. Independent time-use evidence and acceptance criteria must be specified before a confirmatory run. That study has not been executed.

Together, the checks support reproducible descriptions of published occupational activity, visible classification sensitivity and limited comparisons with independent work-activity evidence. They do not support claims about representative workforce adoption, time saved or labour-market effects.