Extension: Who Holds the Pen? Let Specifications, Not Agents, Sign Off
Completion is an authority question. The benchmark separates an agent's completion claim from authoritative state. Across seven models, formal completion claims exceeded official evaluator passes by 28.7 to 37.9 percentage points. This evidence extends the requirement for execution-time artifacts that qualified providers can evaluate. It does not establish that every necessary data boundary or attribution rule was enforced. Those findings are limited to the tested software tasks, models, and evaluators.
Affirmed commentary position. This record extends a source-bound scholarly claim but is not a verbatim paper claim.
Holds when
Current debate
Scholarly basis
Evidence and mapping
Topics
compliancedisclosureinstitutional-design
Provenance
Verify