Australian labour research runs on O*NET without a crosswalk. One now exists, with a warning attached

The problem is arithmetic before it is taxonomy. The best SOC-to-ISCO correspondence the author could find maps 3,349 categories onto 958 O*NET occupations, and none of the matches are one-to-one ^[8]. That works out at roughly 3.5 source categories per O*NET occupation ^[9], and it happens at the first of three joins. ISCO-08 correspondences are published only at unit group level, so OSCA occupations arrive bundled into large groups on the way through ^[10].

The bundling does not run in one direction, which is what stops the chain being fixed by more careful joining. Engineering Technologist is assigned to several ISCO-08 groupings at once ^[7]. The ISCO/ESCO category for sports, recreation and cultural centre managers picks up two O*NET jobs, while O*NET's legislators splits across more than one ISCO/ESCO category ^[11]. On the Australian side, Production Manager (Manufacturing) carries more than one ANZSCO grouping ^[6]. Many-to-many at every step means the person building the table is deciding, not deriving. The author says so directly: analyst judgement is required to make the match fit the use case, and anyone working on trucking occupations, for instance, should confirm by hand that the data has been sensibly assigned ^[12].

That instruction is the part with teeth for anyone citing O*NET-derived numbers in an Australian paper. If the mapping requires judgement, the mapping is a result, with as much claim on the methods section as the model that consumes it.

The pointer to the European Commission's approach is unusual enough to sit with ^[4]. A crosswalk arrives carrying a note from its own author that a better method probably exists elsewhere and has not been reproduced here. Whoever picks up the table inherits that note, whether or not they pass it on.

Then there is decay. OSCA is the new standard and the successor to ANZSCO ^[13], which the author expects will give this crosswalk a short useful life ^[14]. The underlying correspondence tables were sourced from O*NET and the ABS on 21 August 2026 ^[15], so the artefact is pinned to a date and will drift at both ends. The author's own guess at why no official table exists is that OSCA is new and that O*NET-SOC does not match cleanly to ANZSCO or to the intermediate tables anyway ^[17]. The same mismatch turns up whenever datasets use different definitions of industries, administrative boundaries or products ^[18].

Worth noting how the work was made: the code was written largely with Claude, while the write-up was left more or less alone ^[16]. For a data-cleaning and joining exercise, that is the reverse of the usual concern. The thing a reader needs to review is the joins, and the joins came out of a machine. The whole exercise began as groundwork for a 2025 advisory project mapping occupational transition pathways in India ^[1], where the viability of a move between jobs was judged partly on how similar the two occupations were, conditional on geography, wage differentials and education ^[2].