Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
Current evaluations of coding agents may mismeasure performance by conflating action, task, and step levels, revealing a critical flaw in how we assess agent execution.
