Apodex opens a 423-problem discovery benchmark

Apodex introduced TRACES, an executable benchmark built from 423 high-value problems surveyed across 561 industries and 16 sectors, with 20 environments in its initial release. The technical paper says the framework scores six process capabilities—tools, repair, alternatives, coherence, evidence, and scope—and reports a 7% gain over published state of the art on an AAV capsid-design task; the launch brief explains why trajectories and intermediate artifacts are graded alongside outcomes. Researchers evaluating AI’s reproducibility boundary should file this as a benchmark-design shift: an agent’s path to a discovery is becoming part of the result.