Trendora
vien

scale-invariant ground truth

Assess

Techniques

A benchmark design where the correct answer stays constant as dataset scale increases.

Why it's here

Placed in Assess: 1 article(s) of evidence from 1 source(s), led by framework updates, with 0 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.

Evidence (1)

  • 7The New Stack·7/9/2026framework_update
    DevRev releases an enterprise AI agent benchmark

    DevRev introduced an open Enterprise AI Agent Benchmark to better measure how AI agents perform on real enterprise work, especially across large, permissioned data contexts. The first release covers only the L1 and L2 tiers and includes the dataset, evaluation harness, judging criteria, results, and raw traces so others can reproduce the tests.