Mathematical Science Occupations, All Other

Can agents do the work of mathematical science occupations, all other?

We read the evidence at the level of the job family, Data and analytics:

• Task levelPartialOne direct benchmark, near its ceiling. On Spider 2.0-Snow, which asks for multi-step SQL against large business databases, the top agent reaches 96.7% execution accuracy, but the entry names no model. Adjacent spreadsheet tasks agree (83.1% on SpreadsheetBench). No benchmark in our collection tests open-ended analysis, where the right question is itself unclear.
• Project levelPartialMixed, adjacent evidence. On MLE-bench, 75 offline Kaggle competitions, the best agent wins a medal in 64.4% of competitions. The Remote Labor Index includes data analysis projects; across all its projects, the best agent completed 15.8% to a standard a client would accept.