The comparisons evaluate fixed occupational recommendations against observed transition data. SST uses transition shares; EMSI uses the fraction of its original twenty ranked destinations recovered; survey comparisons use supplied transition weights. Mean recovery gives each origin occupation equal weight.
The equal-length comparison uses the smaller available recommendation list for each origin, capped at twenty. The up-to-twenty comparison permits different returned list lengths. Self-transitions are excluded, while observed outcomes outside the candidate population remain in recovery denominators.
Sources, occupational mappings and interpretation · Python and Stata code with aggregate inputs · Model construction and released inputs
Recalculating recovery from released rankings is distinct from reconstructing occupation-task profiles. The latter requires authorized NLx inputs. The public USAJOBS notebooks demonstrate profile construction using a different corpus.